crtc.cs.odu.edu - User contributions [en]

Internal:GPCD

2021-02-26T15:36:58Z

Angelos: /* Papers */

== Introduction ==

=== Protein Cavities ===
Cavities are structural features of proteins and they are often important for drug discovery and design. There are different kinds of cavities, including pockets, voids, tunnels, and channels. The types of cavities are listed in more detail in figure 1.

[[File:proteinCavityTypes.jpg|center|frame| Figure 1: The basic types of protein cavities [1]]]

# '''Void''': A cavity that looks like internal "hole" in a protein with no connection to the surface.
# '''Cleft''': A pocket that looks like a "dent" in the surface of a protein. Can also be found as part of a tunnel or channel.
# '''Invagination''': A pocket that looks like a "cove". It has a small opening on the surface of a protein and leads to a larger interior. Can also be found as part of a tunnel or after a cleft.
# '''Tunnel''': A pocket that has an opening on the surface of a protein and leads to somewhere in the interior. Has no opening on the other side.
# '''Channel''': A cavity that has an opening on one side of the surface of a protein and leads to another opening in another location on the surface of the protein.

Of course, the types of cavities listed in figure 1 are only the basic ones. Combinations of these are also possible as shown in figure 2.

[[File:proteinCavityAdvancedTypes.jpg|center|frame| Figure 2: Combination types of protein cavities [1]]]

=== Protein Cavity Detection Methods ===
There are several methods for detecting protein cavities. There are sphere based methods, grid based methods, grid-and-sphere based methods, surface based methods, grid-and-surface based methods, time-varying methods, GPU based methods, consensus based methods, and finally tessellation based methods. These algorithms are shown in more detail in figure 3. Tessellation based methods mainly rely on Voronoi tessellation. Some methods rely on weighted Voronoi tessellation because of its ability to more accurately represent the volume of atoms. The Voronoi tessellation in these methods is computed from a set of points representing the atomic centers of the underlying protein.

[[File:proteinCavityDetectionAlgorithms.jpg|center|frame| Figure 3: A hierarchy of protein cavity detection algorithms [1]]]

== Literature Review ==

=== MOLE ===

==== Method ====
The MOLE algorithm is specifically designed to identify tunnels and channels. Finding superficial cavities is a secondary concern and does so as part of the surface approximation step. MOLE utilizes the surface obtained by the refinement of the convex hull to find superficial cavities such as clefts and invaginations. In essence, the approximation of a more accurate surface and the detection of these superficial cavities are closely related. As tetrahedra are removed from the convex hull to get a more accurate surface approximation, a parameter called ''interior threshold'' controls how many superficial cavities are found. For example, if the number of tetrahedra removed from a specific part of the molecule exceeds this threshold, the algorithm marks these tetrahedra as part of a superficial cavity.

[[File:moleVoronoiTunnelPath.jpg|center|frame| Figure 4: Shows in 2D the Voronoi diagram (thin lines) and the convex hull (dotted line) of molecule represented by atoms (circles). Each Voronoi edge is evaluated by the cost function (numbers). The thick line represents the optimal path found by the Dijkstra's algorithm from the given starting point (black circle), i.e., centerline of found tunnel. [1]]]

For the more complicated types of cavities, tunnels and channels, MOLE uses the Voronoi diagram to identify them. Start and end points are defined as the centers of the deepest tetrahedra in each cavity. The depth of the tetrahedron is defined as the number of Voronoi edges from the closest boundary tetrahedron (on the surface). Tunnels/channels are computed as the shortest paths between all pairs of start and end points in the same cavity diagram.

[[File:MoleDijkstraWeightFunction.png|center|frame| Figure 5: Shows the weight function used to estimate the weight of each Voronoi edge for Dijkstra's algorihtm. l(e) is the length of the edge, d(e) is the distance of the edge to the closest atom Van der Waals sphere and ϵ is a small number to avoid division by zero [1]]]

==== Complexity ====
* Worst-case: O(N^2 log N), where N is the number of atoms in the molecule.
* Average: O(M log M), where M is the number of vertices in the Voronoi diagram (in the worst case, M = N^2, but in most cases M = O(N)). As a result of the use of the incremental algorithm and Hilbert curve ordering, the complexity of calculating the Delaunay triangulation of most molecular structures is O(N log N). Finally, the complexity of all the remaining steps of the algorithm is at most O(M log M).

=== MolAxis ===

==== Method ====
The MolAxis algorithm relies on Voronoi and Dijkstra's algorithm to find channels, similarly to MOLE. The algorithm consists of the following high-level steps:

# Approximate the molecule using a collection of '''fixed-size spheres'''. The number of balls needed depends on the required quality of the approximation. In the extreme case, where the user is interested in a high resolution, each atom is approximated using a single sphere.
# Compute the '''Voronoi diagram''' of the centers of the fixed-sized spheres. The diagram is intersected with a user-defined bounding sphere. This defines the area of interest.
# Discard the parts of the Voronoi diagram that intersect the molecule.
# Define a '''graph''', using the Voronoi vertices and edges. A weight is given to each edge, which depends on its length and on its distance from the surface of the molecule.
# Define a '''source point (root vertex)'''. The user can either supply a source point or allow MolAxis to automatically choose the source point to be the '''center of the largest cavity in the molecule''' (basically detect channels inside-out).
# Apply a shortest path optimization algorithm ('''Dijktra's algorithm''') on the graph, computing a tree called the "corridor tree".
# The desired channels are paths in the corridor tree. Duplicate channels are avoided by using the split distance parameter. The split distance bounds the distance to the least common ancestor of the channel and all previously reported channels.

[[File:MolAxisVoronoi.png|center|frame| Figure 6: Shows an example of fixed-size circles (light-blue) approximating varying-size circles and the pathway diagram of their centers. The discarded portion of the Voronoi diagram of their centers is depicted using dotted lines [2]. This figure also helps illustrate the usefulness of the Voronoi tessellation in computing pathways.]]

== Protein Energetics ==
=== Metrics ===
# '''Hydrophobicity''': Physical property of a molecule that is seemingly repelled from a mass of water. Hydrophilic molecules are attracted to water. Hydrophobicity plays a major role in protein folding. Hydrophobic molecules also tend to be nonpolar and prefer other neutral molecules and nonpolar solvents. Hydrophobicity is the most widely used metric.
# '''Hydrogen Bonds''': Weak bonds between two molecules resulting from an electrostatic attraction between a proton in one molecule and an electronegative atom in the other. We often see these bonds between oxygen and hydrogen.
# '''Van der Waals Interactions''': Also known as Van der Waals forces, they are forces driven by induced electrical interactions between two or more atoms or molecules that are very close to each other. Van der Waals interaction is the weakest of all intermolecular attractions between molecules. These forces are distance-dependent.
# '''Salt Bridges''': Non-covalent (no sharing of eletrons; electromagnetic) interaction between two ionized sites. It has two components: a hydrogen bond and an electrostatic interaction.
# '''Disulfide Bridges''': Strong bonds that can form between two cysteines (an amino acid). The strength of disulfide bridges helps stabilize a protein. Disulfide bridges are especially common in proteins that are secreted from cells.
# '''Charge''': Plays a role in strength of bonds. A higher charge means that an atom will have stronger forces in an electromagnetic field.
# '''Polarity''': Separation of electric charge leading to a molecule or its chemical groups having an electric dipole moment, with a negatively charged end and a positively charged end. Indicates the existence of polar bonds, which are covalent (sharing of electrons) and are quite strong. About [https://chem.libretexts.org/Bookshelves/Inorganic_Chemistry/Book%3A_Introduction_to_Inorganic_Chemistry/01%3A_Review_of_Chemical_Bonding/1.04%3A_Bond_Polarity_and_Bond_Strength bond polarity].

== Candidate Method: Geometric Protein Cavities Detection ('''GPCD''') ==
The Pipeline contains the following high-level algorithmic steps:

=== (Angelos/Spiros) Read protein ===
# Input: PDB file
# Output: A Polydata of the protein
# Algorithmic Step:
## Using vtkPDBReader, read the PDB file that included the protein. [DONE]
# Challenges:
## Upgrade vtkPDBReader to be able to manage multimodel proteins [DONE]
## Fix VTKPDBReader error of setting wrong atom numbers [DONE]
## Set up vtkPDBReader to utilize [https://sourceforge.net/projects/bodr/ The Blue Obelisk Dataset] for the Covalent radius and Van-der-Waals radius [DONE]
# Ideas that might be used:
## Improve [https://sourceforge.net/projects/bodr/ The Blue Obelisk Dataset] by using: [TODO]
### updated Van-der-Waals radius values provided by Los Alamos National Laboratory
### updated Covalent radius values provided by Cambridge Structural Database

=== (Spiros/Angelos) Remove Het Atoms ===
# Input: A Polydata of the protein
# Output: A Polydata of the protein without Het atoms
# Algorithmic step:
## Using vtkSelectionFilter, we extract all the non-Het atoms. [DONE]

=== (Spiros/Angelos) Sort Atoms By Radius ===
# Input: A Polydata of the protein without Het atoms
# Output: A Polydata of the protein without Het atoms sorted by radius in descending order
# Algorithmic step:
## Using vtkSortFieldData we sort the non-Het atoms by radius in descending order. [DONE]
### '''Note''': Shared memory parallelism can be employed [DONE]

=== (Spiros) Extract protein Chains ===
# Input: A Polydata of the protein without Het atoms sorted by radius in descending order
# Output: A vector of Polydata of the protein chains without Het atoms sorted by radius in descending order
# Algorithmic step:
## Using vtkSelectionFilter, we extract all the protein-chains. [DONE]
### Note: This way we can mesh each chain individually

=== (Spiros/Αngelos) Compute the Solvent Excluded Image ===
# Input: A Polydata of a chain of the protein without Het atoms sorted by radius in descending order, probe radius, spacing
# Output: Solvent Excluded Image of chain
# Algorithmic steps:
## Create the Solvent Accessible image by creating spheres for each atom that their radius is: Van der Waals radius + probe radius. [DONE]
### '''Note''': Shared memory parallelism can be employed [DONE]
## Perform Euclidean Distance Transform on the Solvent Accessible image. [DONE]
### '''Note''': Shared memory parallelism can be employed [DONE]
## Extract Solvent Excluded image from the EDT result by setting: [DONE]
### All pixel values < probe-radius - truncationError[atomType] - paddingDistance = 0
### All pixel values >= probe-radius - truncationError[atomType] - paddingDistance = 1
### '''Note''': Shared memory parallelism can be employed [DONE]
# Challenges:
## Handle the calculation of truncationError[atomType] in a thread-safe way by sorting the atoms by radius in descending order [DONE]
### Truncation error is caused because of the spacing of the image.
## Select a paddingDistance value to help PODM3D produce an accurate surface [DONE]
## Close holes that should not exist (based on other surfaces) [DONE]

=== (Spiros) Compute the Solvent Excluded Surface ('''SES''')===
# Input: Solvent Excluded Image of chain
# Output: Solvent Excluded Surface of chain
# Algorithmic steps:
## Using '''PODM3D''', produce a volumetric mesh of the image. [DONE]
### '''Note''': Shared memory parallelism can be employed [DONE]
## Using '''PODM3D''', extract the surface of the volumetric mesh. [DONE]
### '''Note''': Shared memory parallelism can be employed [DONE]
# Advantages:
## Can control the number of elements on the surface
## the edge lengths bounds are relatively small which helps in the clustering.
# Challenges:
## Using vtkDataSetSurfaceFilter, fix the issue of surface triangles being oriented the wrong way. [DONE]
## Select a default delta value. [DONE]
## Surface needs to be smoothed. [TODO]

=== (Spiros/Angelos) Merge Protein Chain SESs ===
# Input: A vector of the Solvent Excluded Surface of chains
# Output: Solvent Excluded Surface
# Algorithmic step:
## Using vtkAppendPolyData, we merge all the protein-chains surfaces polydata. [DONE]

=== (Angelos/Spiros) Compute ambient occlusion of the SES ===
# Input: Solvent Excluded Surface
# Output: Ambient Occlusion values of surface points, normals of surface points
# Algorithmic step:
## Using IGL Library, which uses [https://www.embree.org/ Embree] and [https://eigen.tuxfamily.org/dox/ Eigen], calculate normals of surface points and then calculate Ambient occlusion values for all the surface points [DONE]
### '''Note''': Shared memory parallelism can be used [DONE]
### '''Note''': GPU parallelism can be used [TODO]
# Ideas that might be used:
## Assign the ambient occlusion value of a triangle by calculating the average of its points. [TODO]
### This can be used for a cell-based clustering method in order to eliminate the noise values caused by the '''PODM3D''''s solvent accessible surface.

===(Spiros) Compute Surface Statistics ===
# Input: Solvent Excluded Surface
# Output: Statistics of Surface Triangles' Edges
# Algoritmic step:
## Compute min/mean/median/max edge lengths of the surface triangles [DONE]
### '''Note''': Shared memory parallelism can be used [DONE]

=== (Angelos) Detect protein cavities ===
# Input: Solvent Excluded Surface, Ambient Occlusion values of surface points, Surface Statistics
# Output: Atomic Cavities
# Algorithmic step:
## Using a modified version of the clustering method '''DBSCAN''', that finds neighbors based on their density '''and''' ambient occlusion values, form clusters that define protein cavities. [DONE]
### '''Note''': shared memory parallelism can be employed, see [https://link.springer.com/article/10.1023/A:1009884809343 A Fast Parallel Clustering Algorithm for Large Spatial Databases PDBSCAN] [TODO]
# Challenges:
## Tune the clustering parameters '''eps / search-radius''' using the surface statistics. [DONE]
## Handle non-deterministic Solvent Excluded Surface of '''PODM3D'''. [TODO]
# Ideas that might be used:
## [https://pdfs.semanticscholar.org/f730/194462fb6bfcd0b90d7f22a960ed58d77199.pdf A Novel Spatial Clustering Algorithm Based on Delaunay Triangulation] [TODO]
## [https://dl.acm.org/doi/abs/10.1145/304181.304187 OPTICS: ordering points to identify the clustering structure] [TODO]

=== (Angelos/Spiros) Classify clusters based on the type of cavity they are: ===
# Input: Solvent Excluded Surface, Atomic Cavities, Atom Types of surface points
# Output: Labeled atomic Cavities
# Algorithm steps:
# Ideas that might be used:
## Ambient occlusion values based classification using mean distance cluster value [TODO]
### '''Note''': Shared memory parallelism can be used [TODO]
## Analyze the shape of clusters to identify what type of cavity they are. [TODO]
### This can potentially be done using machine learning.
### '''Note''': Shared memory parallelism can be used [TODO]

===(Spiros) Classify Atom Types of surface points ===
# Input: Solvent Excluded Surface, a Polydata of the protein without Het atoms sorted by radius in descending order
# Output: Atom Types of surface points
# Algorithmic step:
## Using a KD-tree, for each surface point, find the closest atom: [DONE]
### If the surface point is within the radius of the closest atom, assign the atom type of the closest point to the surface point
### Else, assign '''0''', which means the surface point is part of the re-entrance surface.
### '''Note''': Shared memory parallelism can be used [DONE]
# Challenges:
## Find an ideal offset distance value between the surface points and their closest atoms to improve accuracy. [TODO]
### Idea: use the truncationError[AtomType]
# Ideas that might be used:
## create a different kd-tree for each protein chain [TODO]

=== (Spiros/Angelos) Map Electrostatics Potentials on Surface Points ===
# Input: Solvent Excluded Surface, Electrostatic Potentials Image '''ESP''' of [http://www.poissonboltzmann.org/ APBS], normals of surface points, probe radius, lowerThreshold-upperThreshold
# Output: Electrostatic values of surface points
# Algorithmic steps:
## For each surface point: [DONE]
### Scale the point using normals, and probe radius to the solvent-accessible surface,
### Get the value of the closest pixel from the '''ESP''' to the scaled point:
#### If value > upperThreshold, set electrostatic value of point = upperThreshold
#### Else if, value < lowerThreshold, set electrostatic value of point = lowerThreshold
#### Else get the interpolated value of the scaled points from the image:
##### If interpolated value > upperThreshold, set electrostatic value of point = upperThreshold
##### Else if, interpolated value < lowerThreshold, set electrostatic value of point = lowerThreshold
### '''Note''': Shared memory parallelism can be used [DONE]
# Challenges:
## Create reader of the [https://www.ics.uci.edu/~dock/manuals/apbs/html/user-guide/x2674.html OpenDx] file format that is used as output by APBS [DONE]

=== (Angelos/Spiros) Write Labeled Solvent Surface Surface: ===
# Input: Labeled Solvent Surface Surface
# Output: solventSurfaceSurface.vtp
# Algorithmic step:
## Using a vtkXMLPolyDataWriter, write the output file. [DONE]
# Ideas that might be used:
## Note: a Paraview plugin may be developed for visualization purposes. [TODO]

=== GPCD vs MOLE ===
{| class="wikitable"
|-
! scope="col" | Feature
! scope="col"| GPCD
! scope="col"| MOLE
|-
! scope="row"| Detect Surface Cavities
| style="background-color: green; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Detect Interior Cavities
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Visualize Exact Channel Shape
| style="background-color: red; color: white;" | NO
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Works on Very Large Molecules
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO (very slow)
|-
! scope="row"| Works on Multi-Model Files
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Parallelization
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Relative Performance
| style="background-color: green; color: white;" | VERY FAST
| style="background-color: red; color: white;" | SLOW
|-
! scope="row"| Intuitive Visualization
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| False Positives
| style="background-color: orange; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Quantify Major Structural Features
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|}

=== Resources ===
* [https://blog.kitware.com/what-is-interpolatescalarsbeforemapping-in-vtk/ ParaView vertex colors]

== Resources ==

=== Software ===

* '''[https://www.rbvi.ucsf.edu/chimerax/ UCSF ChimeraX]''': Next-generation molecular visualization program from the Resource for Biocomputing, Visualization, and Informatics (RBVI).
* '''[https://www.ks.uiuc.edu/Research/vmd/ VMD]''': Molecular visualization program for displaying, animating, and analyzing large biomolecular systems using 3D graphics and built-in scripting. Used in conjunction with NAMD for visualizing molecular dynamics simulations.
* '''[https://www.ks.uiuc.edu/Research/namd/ NAMD]''': Parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. Scales to hundreds of cores for typical simulations and beyond 500,000 cores for the largest simulations.
* '''[https://webchem.ncbr.muni.cz/Platform/App/Mole MOLE]''': Universal toolkit for rapid location and characterization of channels and pores in biomacromolecules.
* '''[https://caver.cz/ CAVER]''': Software tool for the analysis and visualization of tunnels and channels in protein structures.
* '''[https://www.rcsb.org/ RCSB PDB Protein Data Bank]''': The largest and most popular database of protein structures.

=== Papers ===

* '''[https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5839519/ Survey: Geometric Protein Cavity Detection Algorithms]''': Paper surveying the current geometric algorithms for protein cavity detection. This is a great resource for understanding this project.
* '''[https://jcheminf.biomedcentral.com/articles/10.1186/1758-2946-5-39 MOLE 2.0]''': Paper describing the MOLE 2.0 protein cavity detection software. Contains a good algorithm that can be used for extracting a good approximation of the surface of a point cloud of atoms.
* '''[https://reader.elsevier.com/reader/sd/pii/S0969212607003759?token=4D0F6192E971D4406E8F7CBA047B0490C8715EBD67579D26427E0E63BD0298BEBF948DB238A28E4EAE2D17EBC6DF4E50 MOLE 1.0]''': Paper describing the MOLE 1.0 algorithm.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/epdf/10.1002/prot.22052 MolAxis]''': Paper describing the MolAxis protein cavity detection software.
* '''[https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002708 CAVER 3.0]''': Paper describing the CAVER 3.0 protein cavity detection software. This algorithm specializes in identifying tunnels.
* '''[https://www.sciencedirect.com/science/article/abs/pii/S0022283606008746 Travel Depth]''': Paper describing the "travel depth" algorithm for protein cavity visualization.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/full/10.1111/cgf.12120 GPU Protein Cavity Detection using Ambient Occlusion]''': Paper describing a GPU-based protein cavity detection algorithm that utilizes Ambient Occlusion.
* '''[https://www.aaai.org/Papers/KDD/1996/KDD96-037.pdf DBSCAN]''': Paper describing the DBSCAN clustering algorithm.
* '''[https://link.springer.com/article/10.1023/A:1009884809343 PDBSCAN]''': Paper describing a parallel version of DBSCAN.
* '''[https://pubmed.ncbi.nlm.nih.gov/31133031/ Coronavirus envelope protein]''': Paper describing research on the envelope protein of coronaviruses.
* '''[https://www.cell.com/biophysj/fulltext/S0006-3495(20)33168-4 COVID-19 virion model]''': Paper describing a detail model of the SARS-CoV-2 virus particle.
* '''[https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7685317/pdf/nihpp-2020.11.19.390187.pdf COVID-19 simulation]''': Paper describing large-scale COVID-19 simulations.

=== Organizations/Groups ===

* '''[https://www.ks.uiuc.edu/ University of Illinois at Urbana-Champaign: Theoretical and Computational Biophysics Group]''': Perhaps the most important group in molecular dynamics. Developers of VMD and NAMD. Ran the simulation for the 200 million atom model of COVID-19.
* '''[https://www.ornl.gov/ Oak Ridge National Laboratory (ORNL)]''': Major national lab with active involvement in drug discovery and design using supercomputers.
* '''[http://rbvi.ucsf.edu/ Resource for Biocomputing, Visualization, and Informatics (RBVI)]''': Developers of UCSF Chimera and UCSF ChimeraX.
* '''[https://ist.ac.at/en/research/edelsbrunner-group/ IST Austria: Edelsbrunner Group]''': Important group in algorithms, computational geometry, and computational topology. Involved with protein cavity detection in the past and potentially involved now. The head of the group, Edelsbrunner created alpha shapes.
* '''[https://www.tacc.utexas.edu/systems/frontera Texas Advanced Computing Center]''': Maintainers of the Frontera supercomputer, the fifth fastest supercomputer in the world (2020). The simulation for the 200 million atom model of COVID-19 was executed on this supercomputer.
* '''[https://webchem.ncbr.muni.cz/ WebChemistry]''': Developed MOLE and many other software tools related to molecular analysis.

=== Important Links ===

* '''[https://www.hpcwire.com/off-the-wire/coronavirus-massive-simulations-completed-on-frontera-supercomputer/ COVID-19 200 million atom NAMD simulation]'''
* '''[https://www.ornl.gov/news/early-research-existing-drug-compounds-supercomputing-could-combat-coronavirus/ ORNL COVID-19 spike drug identification]'''

=== Proteins ===

* '''[https://www.rcsb.org/structure/1jj2 Large Ribosomal Subunit (98,693 atoms)]'''
* '''[https://www.rcsb.org/structure/6VXX COVID-19 Spike Glycoprotein, Closed State (23,694 atoms)]'''
* '''[https://www.rcsb.org/structure/6Y2E COVID-19 Main Protease (5,370 atoms)]'''
* '''[https://www.rcsb.org/structure/6VWW COVID-19 NSP15 Endoribonuclease (24,036 atoms)]'''

== References ==
[1] Simões, T., Lopes, D., Dias, S., Fernandes, F., Pereira, J., Jorge, J., Bajaj, C., & Gomes, A. (2017). Geometric Detection Algorithms for Cavities on Protein Surfaces in Molecular Graphics: A Survey. Computer graphics forum : journal of the European Association for Computer Graphics, 36(8), 643–683. https://doi.org/10.1111/cgf.13158

[2] Yaffe, Fishelovitch, Wolfson, Halperin, & Nussinov. (2008). MolAxis: Efficient and accurate identification of channels in macromolecules. Proteins: Structure, Function, and Bioinformatics, 73(1), 72-86.

[3] Coleman, R. G., & Sharp, K. A. (2006). Travel depth, a new shape descriptor for macromolecules: application to ligand binding. Journal of molecular biology, 362(3), 441–458. https://doi.org/10.1016/j.jmb.2006.07.022

Main Page

2021-01-27T21:17:54Z

Angelos:

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Dynamics===
==== SARS-CoV-2 D614 (original strain) Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S5-D614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 (new strain) Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S6-G614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | DATA SCIENCE ROADMAP TO COMPTON FORM FACTORS OF QUARKS AND GLUONS
|-
|Chrisochoides' Real-Time Computing (CRTC) lab in collaboration with Dr. Charles Hyde (ODU), Dr. Simonetta Liuti (UVA), and Dr. David Heddle (CNU) will host a two-day mini-workshop on the roadmap to Compton Form Factors for efficient evaluations of Monte Carlo Simulations. For more information see: '''[[DATA_SCIENCE_ROADMAP_TO_COMPTON_FORM_FACTORS_OF_QUARKS_AND_GLUONS]]''' [[File:logo.png|250px| center]]
|}

{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
||The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]
|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

People

2021-01-21T21:08:37Z

Angelos: /* Students */

== Principal Investigator ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Nikos_CoS_ODU_2020.png|108px]]
|[https://crtc.cs.odu.edu/People/Nikos-Chrisochoides Dr. Nikos Chrisochoides]

|Exascale Computing: Telescopic Approach for Parallel Anisotropic Mesh Generation and Adaptation

Medical Image Computing: Deformable Registration and Image-to-Mesh Conversion

Multi-Layered Runtime System

|}

== Faculty & Staff ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Chernikov.jpg|108px]]
|Andrey Chernikov Associate Professor

|

|-
|style="background: white"|[[Image:Nadeem.jpg|108px]]
|Tamer Nadeem Associate Professor

|

|-
|style="background: white"|[[Image:ChungHaoChen.png|108px]]
|Chung Hao Chen Assistant Professor

|

|-
|style="background: white"|[[Image:Giannakos.jpg|108px]]
|Michail Giannakos Associate Professor

|

|-
|style="background: white"|[[Image:ThomasKennedy.png|108px]]
|[http://odu.edu/directory/people/t/tkenn013 Thomas Kennedy] Instructor

| High Performance Computing

|-
|style="background: white"|[[Image:Sun.jpg|108px]]
|Jiangwen Sun Assistant Professor

|

|-
|style="background: white"|[[Image:Li.jpg|108px]]
|Weidong Li Visiting Assistant Professor

|

|-
|style="background: white"|[[Image:Liu.jpg|108px]]
|Yixun Liu Adjunct Professor

|

|}

== Students ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Student_Tsolakis.jpg|108px]]
|[http://www.cs.odu.edu/~ctsolakis Christos Tsolakis] Graduate Research Assistant

| Parallel Constrained, Data Refined, and Optimistic for a combination of Delaunay/AFT-based adaptive methods

|-
|style="background: white"|[[Image:Student_Thomadakis.jpg|108px]]
|Polykarpos Thomadakis Graduate Research Assistant

| Parallel Runtime Systems

|-
|style="background: white"|[[Image:REU_KG.jpg|108px]]
|Kevin Garner Graduate Research Assistant

| Parallel Data Refinement using Advancing Front Local Reconnection Mesh Generation (AFLR) Software

|-
|style="background: white"| [[Image:Student_Spiros_Tsalikis.jpg|109px]]
|[https://www.linkedin.com/in/Spiros-Tsalikis/ Spiros Tsalikis] Graduate Research Assistant

| Parallel Mesh Generation, High-Performance Computing

|-
|style="background: white"|[[Image:Student_Angelopoulos.png|108px]]
|[https://www.linkedin.com/in/angelos-angelopoulos-038701162/ Angelos Angelopoulos] Graduate Research Assistant

| Software Engineering, Machine Learning, High-Performance Computing

|-
|style="background: white"|[[Image:Joi.JPG|108px]]
|Joi Best Graduate Research Assistant

| Image to Mesh Conversion
|}

== External Collaborators ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Darmofal_David.jpg|108px]]
|[https://darmofal.mit.edu/ David Darmofal]

| Parallel Error-based metrics

Parallel Adaptive Mesh Refinement

|-
|style="background: white"|[[Image:Marcum_David.jpg|108px]]
| [http://www.me.msstate.edu/faculty/marcum/marcum.html David Marcum]

| Parallel Anisotropic AFT-based Mesh Generation and Adaption

Interactions with Geometry

|-
|style="background: white"|[[Image:Collaborator_Jiao.jpg|108px]]
| [http://www.ams.sunysb.edu/~jiao/ Prof. Jim Jiao]

| Scalable Abstractions for Geometry

|-
|style="background: white"|[[Image:Student_Drakopoulos.jpg|108px]]
|Fotis Drakopoulos

| Delaunay and AFT-based mesh generation and their speculative parallel implementations

|-
|style="background: white"|[[Image:Collaborator_Kot.jpg|108px]]
| [http://web-test.ncsa.illinois.edu/AboutUs/People/contact.php?id=kot Dr. Andriy Kot]

| Parallel Runtime Software Systems

|-
|style="background: white"|[[Image:Visitor_Nikolopoulos.jpg|108px]]
| [http://www.cs.qub.ac.uk/~D.Nikolopoulos/ Prof. Dimitris Nikolopoulos]

| Parallel Power Aware Runtime Systems

Programming Languages

|}

== Visitors ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Antonopoulos.jpg|108px]]
|Christos Antonopoulos Adjunct Senior Research Associate

|

|-
|style="background: white"|[[Image:Schneipp.jpg|108px]]
|Hannes Schneipp Assistant Professor

|

|-
|style="background: white"|[[Image:Zhang.jpg|108px]]
|Jessica Zhang Professor

|

|-
|style="background: white"|[[Image:Lesnick.jpg|108px]]
|Jim Lesnick RMG Neurosurgeon

|

|-
|style="background: white"|[[Image:Han.jpg|108px]]
|Joseph Han Associate Professor

|

|-
|style="background: white"|[[Image:Behr.jpg|108px]]
|Marek Behr Professor

|

|-
|style="background: white"|[[Image:Clatz.jpg|108px]]
|Olivier Clatz Research Fellow

|

|-
|style="background: white"|[[Image:Shontz.jpg|108px]]
|Suzanne M. Shontz Professor

|

|-
|style="background: white"|[[Image:Papademitris.jpg|108px]]
|Xenios Papademitris Assistant Professor

|

|}

== Alumni ==

{| class="wikitable"

|-
|style="background: white"|[[Image:Ahmed.png|108px]]
|Ahmed Fakhry Research Assistant

|

|-
|style="background: white"|[[Image:Williams.jpg|108px]]
|Alex Williams Visiting High School Research Assistant

|

|-
|style="background: white"|[[Image:REU_Alexis.jpg|108px]]
|Alexis Brueggeman REU Student

|

|-
|style="background: white"|[[Image:Smith.jpg|108px]]
|Alison Smith Undergraduate Research Assistant

|

|-
|style="background: white"|[[Image:McKnight.jpg|108px]]
|Andrew McKnight REU Student

|

|-
|style="background: white"|[[Image:Fedorov.jpg|108px]]
|Andriy Fedorov Research Assistant

|

|-
|style="background: white"|[[Image:Holinka.jpg|108px]]
|Brian Holinka Research Assistant

|

|-
|style="background: white"|[[Image:Verma.jpg|108px]]
|Chaman Verma Research Assistant

|

|-
|style="background: white"|[[Image:Feng.jpg|108px]]
|Daming Feng Graduate Research Assistant

|

|-
|style="background: white"|[[Image:Nave.jpg|108px]]
|Demian Nave Research Assistant

|

|-
|style="background: white"|[[Image:Kenzor.png|108px]]
|Denis Kenzor Research Assistant

|

|-
|style="background: white"|[[Image:Adam.jpg|108px]]
|Eleni Adam Graduate Research Assistant

|

|-
|style="background: white"|[[Image:Billet.jpg|108px]]
|Eric Billet Research Assistant

|

|-
|style="background: white"|[[Image:Zagaris.jpg|108px]]
|George Zagaris Research Assistant

|

|-
|style="background: white"|
|Hugh Carson MIT, Research Assistant

|

|-
|style="background: white"|[[Image:Xu.jpg|108px]]
|Jing Xu Research Assistant

|

|-
|style="background: white"|[[Image:Karthik.jpg|108px]]
|Karthik Navuluri PhD Student

|

|-
|style="background: white"|[[Image:Barker.jpg|108px]]
|Kevin Barker Research Assistant

|

|-
|style="background: white"|[[Image:Almahallawy.jpg|108px]]
|Khaled Almahallawy Research Assistant

|

|-
|style="background: white"|[[Image:Student_Kyriaki_Kavazidi.jpg|108px]]
|Kyriaki Rafailia Kavazidi

| Medical Student

|-
|style="background: white"|[[Image:Linardakis.jpg|108px]]
|Leonidas Linardakis Research Associate

|

|-
|style="background: white"|[[Image:Hines.png|108px]]
|Matt Hines REU Student

|

|-
|style="background: white"|[[Image:Trotta.jpg|108px]]
|Michael Trotta REU Student

|

|-
|style="background: white"|[[Image:Weissberger.jpg|108px]]
|Michael Weissberger Research Assistant

|

|-
|style="background: white"|[[Image:MirandaSmith.png|108px]]
|Miranda Smith REU Student

|

|-
|style="background: white"|[[Image:Foteinos.jpg|108px]]
|Panagiotis Foteinos Research Assistant

|

|-
|style="background: white"|[[Image:Student_Caplan.jpg|108px]]
|Philip Caplan MIT, Research Assistant

|

|-
|style="background: white"|[[Image:Rongjian.png|108px]]
|Rongjian Li Research Assistant

|

|-
|style="background: white"|[[Image:Ji.jpg|108px]]
|Shuiwang Ji Assistant Professor

|

|-
|style="background: white"|[[Image:Folsom.jpg|108px]]
|Sydney Folsom Visiting High School Research Assistant

|

|-
|style="background: white"|[[Image:Zeng.png|108px]]
|Tao Zeng Research Assistant

|

|-
|style="background: white"|[[Image:WenluZhang.jpg|108px]]
|Wenlu Zhang PhD Student

|

|}

CNF:ML Data Denoise

2020-12-03T21:57:45Z

Angelos: /* Z Plane Quantization */

This page will be updated about the machine learning data de-noising effort.
= Problem =
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.

= Goal =
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.

= Results =
Hardware specs:

'''Tesla V100 GPU'''

== 3D Input ==
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy.png]]

Example correct output:

[[File:cnf_ml_denoise_3d.png]]

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
|-
| 20
| 0.000064
| 0.00007
| 7822
| 0.00087
| [[File:cnf_ml_cnn_v2_train_val_loss_3d.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised_3d.png]]
|}

=== Different Noise ===
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy2.png]]

Example correct output:

[[File:cnf_ml_denoise_3d2.png]]

{| class="wikitable
! Noisy
! Correct
! Denoised
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex1.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex2.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex3.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex4.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex5.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex6.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_denoise.png]]
|}

[[File:cnf_ml_denoise_3d_noisy2_hits_hist.png]]
[[File:cnf_ml_denoise_3d_noisy2_noise_hist.png]]

=== Z Plane Quantization ===

==== Quantization Process ====
# Establish an array of Z ranges (must have an odd size)
# The number of added planes between each base plane is equal to the truncated half size of the Z range array
# Iterate hitlist activations
## Get the plane, ring, and pad numbers
## Offset the plane number to ignore added planes
## Offset the plane number by the difference of the center index of the Z ranges array from the index of the range the Z value belongs

{|
! style="text-align: center; border-style: solid solid none solid;" | No Quantization
(10 planes)
! style="text-align: center; border-style: solid solid none none;"| Quantization
(21 planes)
! style="text-align: center; border-style: solid solid none none;"| Quantization
(32 planes)
! style="text-align: center; border-style: solid solid none none;"| Quantization
(43 planes)
|-
|style="text-align: center; border-style: none solid solid solid;"| Z Ranges:
{}
|style="text-align: center; border-style: none solid solid solid;"| Z Ranges:
{(-20, -1.5), (-1.5, 1.5), (1.5, 21)}
|style="text-align: center; border-style: none solid solid none;"| Z Ranges:
{(-20, -10), (-10, -1.5), (-1.5, 1.5), (1.5, 10), (10, 21)}
|style="text-align: center; border-style: none solid solid none;"| Z Ranges:
{(-20, -15), (-15, -10), (-10, -1.5), (-1.5, 1.5), (1.5, 10), (10, 15), (15, 21)}
|-
| style="text-align: center; vertical-align:top;"| [[File:CNF_3D_NO_QUANTIZATION.png]]
| style="text-align: center; vertical-align:top;"| [[File:CNF_3D_QUANTIZATION_21.png]]
| style="text-align: center; vertical-align:top;"| [[File:CNF_3D_QUANTIZATION_32.png]]
| style="text-align: center; vertical-align:top;"| [[File:CNF_3D_QUANTIZATION_43.png]]
|}

== 2D Input ==
Example noisy input:

[[File:cnf_ml_noisy_track.png|480px]]

Example correct input:

[[File:cnf_ml_correct_track.png|480px]]

'''Both the MLP and CNN networks used relu activation for all hidden layers and sigmoid for the output. Binary cross-entropy was used as the loss function and Adam as the optimizer.'''

=== Multilayer Perceptrons ===
[[File:mlp_ae_model.png]]

{| class="wikitable
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 0.0018
| 0.0219
| 298
| 0.000182
| [[File:mlp_ae_loss.png|480px]]
| [[File:cnf_ml_mlp_denoised_track.png|480px]]
| [[File:cnf_ml_mlp_hist.png|480px]]
| [[File:cnf_ml_mlp_noise_hist.png|480px]]
|}

=== Convolutional Networks ===
[[File:cnn_ae_model.png]]

'''Train and test on single track'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0024
| 0.0024
| 238
| 0.000123
| [[File:cnn_ae_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_single_track.png|480px]]
|-
| 100
| 0.0023
| 0.0025
| 471.2
| 0.0001
| [[File:cnn_ae_loss_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_100e_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_100e_10_single_track.png|480px]]
|}

'''Train and test on single track V2'''

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 12
| 0.00048
| 0.00085
| 204
| 0.00024
| [[File:cnf_ml_cnn_v2_train_val_loss.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised.png|480px]]
| [[File:cnf_ml_cnn_v2_hits_histogram.png|480px]]
| [[File:cnf_ml_cnn_v2_noise_histogram.png|480px]]
|}

Per segment histograms:

[[File:cnf_ml_cnn_v2_hits_histogram_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg5.png|380px]]

[[File:0.noise_histogram.png|480px]]
[[File:1.noise_histogram.png|480px]]
[[File:2.noise_histogram.png|480px]]
[[File:3.noise_histogram.png|480px]]
[[File:4.noise_histogram.png|480px]]
[[File:5.noise_histogram.png|480px]]

'''Train and test on multiple tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0029
| 0.0030
| 683
| 0.0002
| [[File:cnn_ae_train_val_mult_loss_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_v2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg5.png|380px]]

[[File:0.noise_histogram_mult.png|480px]]
[[File:1.noise_histogram_mult.png|480px]]
[[File:2.noise_histogram_mult.png|480px]]
[[File:3.noise_histogram_mult.png|480px]]
[[File:4.noise_histogram_mult.png|480px]]
[[File:5.noise_histogram_mult.png|480px]]

'''Train on single and test on multiple tracks'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.027
| 0.00012
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track.png|480px]]
|-
| 100
| 0.024
| 0.000125
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track_100e.png|480px]]
|}

'''Train and test on multiple tracks'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.00537
| 0.00505
|
| 0.0001
| [[File:cnn_ae_train_val_mult_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track.png|480px]]
|}

'''Train and test on multiple tracks with more kernels'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0041
| 0.0040
| 943
| 0.0001
| [[File:cnn_ae_train_val_mult_loss_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_more.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks'''

Correct Track:

[[File:cnf_ml_cnn_correct_mult2_3_track.png|480px]]

Noisy Input Track:

[[File:cnf_ml_cnn_correct_mult2_3_track_noisy.png|480px]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0078
|
| [[File:cnf_ml_cnn_denoised_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0010
| 0.0007
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg5.png|380px]]

[[File:0.noise_histogram_mult23.png|480px]]
[[File:1.noise_histogram_mult23.png|480px]]
[[File:2.noise_histogram_mult23.png|480px]]
[[File:3.noise_histogram_mult23.png|480px]]
[[File:4.noise_histogram_mult23.png|480px]]
[[File:5.noise_histogram_mult23.png|480px]]

'''Train on new 2-3 tracks and test on multitrack V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.003
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg5.png|380px]]

[[File:0.noise_histogram_multitrack_new.png|480px]]
[[File:1.noise_histogram_multitrack_new.png|480px]]
[[File:2.noise_histogram_multitrack_new.png|480px]]
[[File:3.noise_histogram_multitrack_new.png|480px]]
[[File:4.noise_histogram_multitrack_new.png|480px]]
[[File:5.noise_histogram_multitrack_new.png|480px]]

'''Train on new 2-3 tracks and test on old 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.0045
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg5.png|380px]]

[[File:0.noise_histogram_mult23_new.png|480px]]
[[File:1.noise_histogram_mult23_new.png|480px]]
[[File:2.noise_histogram_mult23_new.png|480px]]
[[File:3.noise_histogram_mult23_new.png|480px]]
[[File:4.noise_histogram_mult23_new.png|480px]]
[[File:5.noise_histogram_mult23_new.png|480px]]

'''Train on multiple and test on 2-3 tracks with more kernels'''

[[File:cnn_ae_model_more_kernels.png]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time

! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0075
| 0.0008

| [[File:cnf_ml_cnn_denoised_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_more.png|480px]]
|}

=== Examples ===

{| class="wikitable
! Noisy
! Correct
! MLP
! CNN
! CNN V2
! CNN V2 new
|-
| [[File:ae_noisy_109.png|360px]]
| [[File:ae_correct_109.png|360px]]
| [[File:ae_mlp_denoised_109.png|360px]]
| [[File:ae_cnn_denoised_109.png|360px]]
|-
| [[File:ae_noisy_1019.png|360px]]
| [[File:ae_correct_1019.png|360px]]
| [[File:ae_mlp_denoised_1019.png|360px]]
| [[File:ae_cnn_denoised_1019.png|360px]]
|-
| [[File:ae_noisy_312.png|360px]]
| [[File:ae_correct_312.png|360px]]
| [[File:ae_mlp_denoised_312.png|360px]]
| [[File:ae_cnn_denoised_312.png|360px]]
|-
| [[File:ae_noisy_666.png|360px]]
| [[File:ae_correct_666.png|360px]]
| [[File:ae_mlp_denoised_666.png|360px]]
| [[File:ae_cnn_denoised_666.png|360px]]
|-
| [[File:ae_noisy_1.png|360px]]
| [[File:ae_correct_1.png|360px]]
| [[File:ae_mlp_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1_V2.png|360px]]
| [[File:ae_cnn_denoised_1_V2_new.png|360px]]
|-
| [[File:ae_noisy_2.png|360px]]
| [[File:ae_correct_2.png|360px]]
| [[File:ae_mlp_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2_V2.png|360px]]
| [[File:ae_cnn_denoised_2_V2_new.png|360px]]
|-
| [[File:ae_noisy_3.png|360px]]
| [[File:ae_correct_3.png|360px]]
| [[File:ae_mlp_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3_V2.png|360px]]
| [[File:ae_cnn_denoised_3_V2_new.png|360px]]
|-
| [[File:ae_noisy_1m.png|360px]]
| [[File:ae_correct_1m.png|360px]]
| [[File:ae_mlp_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m_V2.png|360px]]
| [[File:ae_cnn_denoised_1m_V2_new.png|360px]]
|-
| [[File:ae_noisy_2m.png|360px]]
| [[File:ae_correct_2m.png|360px]]
| [[File:ae_mlp_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m_V2.png|360px]]
| [[File:ae_cnn_denoised_2m_V2_new.png|360px]]
|-
| [[File:ae_noisy_3m.png|360px]]
| [[File:ae_correct_3m.png|360px]]
| [[File:ae_mlp_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m_V2.png|360px]]
| [[File:ae_cnn_denoised_3m_V2_new.png|360px]]
|-
| [[File:ae_noisy_4m.png|360px]]
| [[File:ae_correct_4m.png|360px]]
| [[File:ae_mlp_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m_V2.png|360px]]
| [[File:ae_cnn_denoised_4m_V2_new.png|360px]]
|}

=== Work Plan ===

* create a histogram comparing how many points in the reconstructed image match points in the ground truth
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation
* create the efficiency and noise plots for the multi-track data sample reconstruction.
* train network with multi-track data and evaluate performance on multi-track testing sample.
* put on the web site architecture of the network and try with a little deeper architecture (4032->2016->1008->336 -> 224 -> 336 -> 1008 -> 2016 -> 4032)

=== Work Plan (November 17 ===

* Provide workflow for testing, given input image write out the truth and AI inferred image
* Write software to create an image from AI inferred image and Truth image to reconstruct better truth, by filling in missing pieces in AI produced image from truth image when the points are close
* Evaluate efficiency of reconstructing the truth image and noisiness of resulting image. Also make graph now much noise was removed from original image.
* Start working with Hall-C data. much bigger input and evaluate efficiency of cleaning noise
* at the first stage noise has to be added to the Hall-C data for training, it can start with one track sample, then 250 track sample.

= Resources =
https://userweb.jlab.org/~gavalian/Documents/NN_Vary_Width_Depth%20(Nguyen).pdf

https://userweb.jlab.org/~gavalian/ML/2020/AutoEncoder/

File:CNF 3D QUANTIZATION 43.png

2020-12-03T21:55:27Z

Angelos:

File:CNF 3D QUANTIZATION 32.png

2020-12-03T21:54:19Z

Angelos:

File:CNF 3D QUANTIZATION 21.png

2020-12-03T21:53:43Z

Angelos:

CNF:ML Data Denoise

2020-12-03T00:25:13Z

Angelos: /* Z Plane Quantization */

This page will be updated about the machine learning data de-noising effort.
= Problem =
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.

= Goal =
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.

= Results =
Hardware specs:

'''Tesla V100 GPU'''

== 3D Input ==
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy.png]]

Example correct output:

[[File:cnf_ml_denoise_3d.png]]

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
|-
| 20
| 0.000064
| 0.00007
| 7822
| 0.00087
| [[File:cnf_ml_cnn_v2_train_val_loss_3d.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised_3d.png]]
|}

=== Different Noise ===
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy2.png]]

Example correct output:

[[File:cnf_ml_denoise_3d2.png]]

{| class="wikitable
! Noisy
! Correct
! Denoised
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex1.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex2.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex3.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex4.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex5.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex6.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_denoise.png]]
|}

[[File:cnf_ml_denoise_3d_noisy2_hits_hist.png]]
[[File:cnf_ml_denoise_3d_noisy2_noise_hist.png]]

=== Z Plane Quantization ===

==== Quantization Process ====
# Establish an array of Z ranges (must have an odd size)
# The number of added planes between each base plane is equal to the truncated half size of the Z range array
# Iterate hitlist activations
## Get the plane, ring, and pad numbers
## Offset the plane number to ignore added planes
## Offset the plane number by the difference of the center index of the Z ranges array from the index of the range the Z value belongs

'''No Quantization (10 planes):'''

[[File:CNF_3D_NO_QUANTIZATION.png]]

'''Quantization (24 planes):'''

''Z Ranges:'' [(-20, -15), (-15, -5), (-5, 5), (5, 15), (15, 21)]

[[File:CNF_3D_QUANTIZATION_24.png]]

'''Quantization (36 planes):'''

''Z Ranges:'' [(-20, -15), (-15, -10), (-10, -5), (-5, 5), (5, 10), (10, 15), (15, 21)]

[[File:CNF_3D_QUANTIZATION_36.png]]

== 2D Input ==
Example noisy input:

[[File:cnf_ml_noisy_track.png|480px]]

Example correct input:

[[File:cnf_ml_correct_track.png|480px]]

'''Both the MLP and CNN networks used relu activation for all hidden layers and sigmoid for the output. Binary cross-entropy was used as the loss function and Adam as the optimizer.'''

=== Multilayer Perceptrons ===
[[File:mlp_ae_model.png]]

{| class="wikitable
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 0.0018
| 0.0219
| 298
| 0.000182
| [[File:mlp_ae_loss.png|480px]]
| [[File:cnf_ml_mlp_denoised_track.png|480px]]
| [[File:cnf_ml_mlp_hist.png|480px]]
| [[File:cnf_ml_mlp_noise_hist.png|480px]]
|}

=== Convolutional Networks ===
[[File:cnn_ae_model.png]]

'''Train and test on single track'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0024
| 0.0024
| 238
| 0.000123
| [[File:cnn_ae_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_single_track.png|480px]]
|-
| 100
| 0.0023
| 0.0025
| 471.2
| 0.0001
| [[File:cnn_ae_loss_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_100e_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_100e_10_single_track.png|480px]]
|}

'''Train and test on single track V2'''

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 12
| 0.00048
| 0.00085
| 204
| 0.00024
| [[File:cnf_ml_cnn_v2_train_val_loss.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised.png|480px]]
| [[File:cnf_ml_cnn_v2_hits_histogram.png|480px]]
| [[File:cnf_ml_cnn_v2_noise_histogram.png|480px]]
|}

Per segment histograms:

[[File:cnf_ml_cnn_v2_hits_histogram_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg5.png|380px]]

[[File:0.noise_histogram.png|480px]]
[[File:1.noise_histogram.png|480px]]
[[File:2.noise_histogram.png|480px]]
[[File:3.noise_histogram.png|480px]]
[[File:4.noise_histogram.png|480px]]
[[File:5.noise_histogram.png|480px]]

'''Train and test on multiple tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0029
| 0.0030
| 683
| 0.0002
| [[File:cnn_ae_train_val_mult_loss_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_v2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg5.png|380px]]

[[File:0.noise_histogram_mult.png|480px]]
[[File:1.noise_histogram_mult.png|480px]]
[[File:2.noise_histogram_mult.png|480px]]
[[File:3.noise_histogram_mult.png|480px]]
[[File:4.noise_histogram_mult.png|480px]]
[[File:5.noise_histogram_mult.png|480px]]

'''Train on single and test on multiple tracks'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.027
| 0.00012
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track.png|480px]]
|-
| 100
| 0.024
| 0.000125
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track_100e.png|480px]]
|}

'''Train and test on multiple tracks'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.00537
| 0.00505
|
| 0.0001
| [[File:cnn_ae_train_val_mult_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track.png|480px]]
|}

'''Train and test on multiple tracks with more kernels'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0041
| 0.0040
| 943
| 0.0001
| [[File:cnn_ae_train_val_mult_loss_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_more.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks'''

Correct Track:

[[File:cnf_ml_cnn_correct_mult2_3_track.png|480px]]

Noisy Input Track:

[[File:cnf_ml_cnn_correct_mult2_3_track_noisy.png|480px]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0078
|
| [[File:cnf_ml_cnn_denoised_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0010
| 0.0007
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg5.png|380px]]

[[File:0.noise_histogram_mult23.png|480px]]
[[File:1.noise_histogram_mult23.png|480px]]
[[File:2.noise_histogram_mult23.png|480px]]
[[File:3.noise_histogram_mult23.png|480px]]
[[File:4.noise_histogram_mult23.png|480px]]
[[File:5.noise_histogram_mult23.png|480px]]

'''Train on new 2-3 tracks and test on multitrack V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.003
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg5.png|380px]]

[[File:0.noise_histogram_multitrack_new.png|480px]]
[[File:1.noise_histogram_multitrack_new.png|480px]]
[[File:2.noise_histogram_multitrack_new.png|480px]]
[[File:3.noise_histogram_multitrack_new.png|480px]]
[[File:4.noise_histogram_multitrack_new.png|480px]]
[[File:5.noise_histogram_multitrack_new.png|480px]]

'''Train on new 2-3 tracks and test on old 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.0045
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg5.png|380px]]

[[File:0.noise_histogram_mult23_new.png|480px]]
[[File:1.noise_histogram_mult23_new.png|480px]]
[[File:2.noise_histogram_mult23_new.png|480px]]
[[File:3.noise_histogram_mult23_new.png|480px]]
[[File:4.noise_histogram_mult23_new.png|480px]]
[[File:5.noise_histogram_mult23_new.png|480px]]

'''Train on multiple and test on 2-3 tracks with more kernels'''

[[File:cnn_ae_model_more_kernels.png]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time

! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0075
| 0.0008

| [[File:cnf_ml_cnn_denoised_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_more.png|480px]]
|}

=== Examples ===

{| class="wikitable
! Noisy
! Correct
! MLP
! CNN
! CNN V2
! CNN V2 new
|-
| [[File:ae_noisy_109.png|360px]]
| [[File:ae_correct_109.png|360px]]
| [[File:ae_mlp_denoised_109.png|360px]]
| [[File:ae_cnn_denoised_109.png|360px]]
|-
| [[File:ae_noisy_1019.png|360px]]
| [[File:ae_correct_1019.png|360px]]
| [[File:ae_mlp_denoised_1019.png|360px]]
| [[File:ae_cnn_denoised_1019.png|360px]]
|-
| [[File:ae_noisy_312.png|360px]]
| [[File:ae_correct_312.png|360px]]
| [[File:ae_mlp_denoised_312.png|360px]]
| [[File:ae_cnn_denoised_312.png|360px]]
|-
| [[File:ae_noisy_666.png|360px]]
| [[File:ae_correct_666.png|360px]]
| [[File:ae_mlp_denoised_666.png|360px]]
| [[File:ae_cnn_denoised_666.png|360px]]
|-
| [[File:ae_noisy_1.png|360px]]
| [[File:ae_correct_1.png|360px]]
| [[File:ae_mlp_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1_V2.png|360px]]
| [[File:ae_cnn_denoised_1_V2_new.png|360px]]
|-
| [[File:ae_noisy_2.png|360px]]
| [[File:ae_correct_2.png|360px]]
| [[File:ae_mlp_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2_V2.png|360px]]
| [[File:ae_cnn_denoised_2_V2_new.png|360px]]
|-
| [[File:ae_noisy_3.png|360px]]
| [[File:ae_correct_3.png|360px]]
| [[File:ae_mlp_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3_V2.png|360px]]
| [[File:ae_cnn_denoised_3_V2_new.png|360px]]
|-
| [[File:ae_noisy_1m.png|360px]]
| [[File:ae_correct_1m.png|360px]]
| [[File:ae_mlp_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m_V2.png|360px]]
| [[File:ae_cnn_denoised_1m_V2_new.png|360px]]
|-
| [[File:ae_noisy_2m.png|360px]]
| [[File:ae_correct_2m.png|360px]]
| [[File:ae_mlp_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m_V2.png|360px]]
| [[File:ae_cnn_denoised_2m_V2_new.png|360px]]
|-
| [[File:ae_noisy_3m.png|360px]]
| [[File:ae_correct_3m.png|360px]]
| [[File:ae_mlp_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m_V2.png|360px]]
| [[File:ae_cnn_denoised_3m_V2_new.png|360px]]
|-
| [[File:ae_noisy_4m.png|360px]]
| [[File:ae_correct_4m.png|360px]]
| [[File:ae_mlp_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m_V2.png|360px]]
| [[File:ae_cnn_denoised_4m_V2_new.png|360px]]
|}

=== Work Plan ===

* create a histogram comparing how many points in the reconstructed image match points in the ground truth
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation
* create the efficiency and noise plots for the multi-track data sample reconstruction.
* train network with multi-track data and evaluate performance on multi-track testing sample.
* put on the web site architecture of the network and try with a little deeper architecture (4032->2016->1008->336 -> 224 -> 336 -> 1008 -> 2016 -> 4032)

=== Work Plan (November 17 ===

* Provide workflow for testing, given input image write out the truth and AI inferred image
* Write software to create an image from AI inferred image and Truth image to reconstruct better truth, by filling in missing pieces in AI produced image from truth image when the points are close
* Evaluate efficiency of reconstructing the truth image and noisiness of resulting image. Also make graph now much noise was removed from original image.
* Start working with Hall-C data. much bigger input and evaluate efficiency of cleaning noise
* at the first stage noise has to be added to the Hall-C data for training, it can start with one track sample, then 250 track sample.

= Resources =
https://userweb.jlab.org/~gavalian/Documents/NN_Vary_Width_Depth%20(Nguyen).pdf

https://userweb.jlab.org/~gavalian/ML/2020/AutoEncoder/

CNF:ML Data Denoise

2020-12-02T20:35:03Z

Angelos:

This page will be updated about the machine learning data de-noising effort.
= Problem =
Traditional tracking algorithms are computationally intensive, especially for high luminosity experiments with multi-track final states where all combinations of segments in drift chambers have to be considered for firing best track candidates. At high luminosity the number of random segments (unrelated to the tracks) are increasing and as a result the number of possible combinations also increases, making the whole process longer.

= Goal =
Use Deep Learning Autoencoders that are trained on noisy tracks to produce the original track without the noise.

= Results =
Hardware specs:

'''Tesla V100 GPU'''

== 3D Input ==
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy.png]]

Example correct output:

[[File:cnf_ml_denoise_3d.png]]

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
|-
| 20
| 0.000064
| 0.00007
| 7822
| 0.00087
| [[File:cnf_ml_cnn_v2_train_val_loss_3d.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised_3d.png]]
|}

=== Different Noise ===
Example noisy input:

[[File:cnf_ml_denoise_3d_noisy2.png]]

Example correct output:

[[File:cnf_ml_denoise_3d2.png]]

{| class="wikitable
! Noisy
! Correct
! Denoised
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex1.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex1_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex2.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex2_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex3.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex3_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex4.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex4_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex5.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex5_denoise.png]]
|-
|[[File:cnf_ml_denoise_3d_noisy2_ex6.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_cor.png]]
|[[File:cnf_ml_denoise_3d_noisy2_ex6_denoise.png]]
|}

[[File:cnf_ml_denoise_3d_noisy2_hits_hist.png]]
[[File:cnf_ml_denoise_3d_noisy2_noise_hist.png]]

=== Z Plane Quantization ===

No Quantization (10 planes):

[[File:CNF_3D_NO_QUANTIZATION.png]]

Quantization (24 planes):

[[File:CNF_3D_QUANTIZATION_24.png]]

Quantization (36 planes):

[[File:CNF_3D_QUANTIZATION_36.png]]

== 2D Input ==
Example noisy input:

[[File:cnf_ml_noisy_track.png|480px]]

Example correct input:

[[File:cnf_ml_correct_track.png|480px]]

'''Both the MLP and CNN networks used relu activation for all hidden layers and sigmoid for the output. Binary cross-entropy was used as the loss function and Adam as the optimizer.'''

=== Multilayer Perceptrons ===
[[File:mlp_ae_model.png]]

{| class="wikitable
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 0.0018
| 0.0219
| 298
| 0.000182
| [[File:mlp_ae_loss.png|480px]]
| [[File:cnf_ml_mlp_denoised_track.png|480px]]
| [[File:cnf_ml_mlp_hist.png|480px]]
| [[File:cnf_ml_mlp_noise_hist.png|480px]]
|}

=== Convolutional Networks ===
[[File:cnn_ae_model.png]]

'''Train and test on single track'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0024
| 0.0024
| 238
| 0.000123
| [[File:cnn_ae_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_single_track.png|480px]]
|-
| 100
| 0.0023
| 0.0025
| 471.2
| 0.0001
| [[File:cnn_ae_loss_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_100e_10_single_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_100e_10_single_track.png|480px]]
|}

'''Train and test on single track V2'''

[[File:cnf_ml_cnnv2_arch.png]]
4x6 Kernels

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 12
| 0.00048
| 0.00085
| 204
| 0.00024
| [[File:cnf_ml_cnn_v2_train_val_loss.png|480px]]
| [[File:cnf_ml_cnn_v2_cnn_denoised.png|480px]]
| [[File:cnf_ml_cnn_v2_hits_histogram.png|480px]]
| [[File:cnf_ml_cnn_v2_noise_histogram.png|480px]]
|}

Per segment histograms:

[[File:cnf_ml_cnn_v2_hits_histogram_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_seg5.png|380px]]

[[File:0.noise_histogram.png|480px]]
[[File:1.noise_histogram.png|480px]]
[[File:2.noise_histogram.png|480px]]
[[File:3.noise_histogram.png|480px]]
[[File:4.noise_histogram.png|480px]]
[[File:5.noise_histogram.png|480px]]

'''Train and test on multiple tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0029
| 0.0030
| 683
| 0.0002
| [[File:cnn_ae_train_val_mult_loss_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_v2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_v2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult_seg5.png|380px]]

[[File:0.noise_histogram_mult.png|480px]]
[[File:1.noise_histogram_mult.png|480px]]
[[File:2.noise_histogram_mult.png|480px]]
[[File:3.noise_histogram_mult.png|480px]]
[[File:4.noise_histogram_mult.png|480px]]
[[File:5.noise_histogram_mult.png|480px]]

'''Train on single and test on multiple tracks'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.027
| 0.00012
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track.png|480px]]
|-
| 100
| 0.024
| 0.000125
|
| [[File:cnf_ml_cnn_denoised_histogram_10_train_single_test_mult_track_100e.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_single_test_mult_track_100e.png|480px]]
|}

'''Train and test on multiple tracks'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.00537
| 0.00505
|
| 0.0001
| [[File:cnn_ae_train_val_mult_loss.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track.png|480px]]
|}

'''Train and test on multiple tracks with more kernels'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0041
| 0.0040
| 943
| 0.0001
| [[File:cnn_ae_train_val_mult_loss_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult_track_more.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks'''

Correct Track:

[[File:cnf_ml_cnn_correct_mult2_3_track.png|480px]]

Noisy Input Track:

[[File:cnf_ml_cnn_correct_mult2_3_track_noisy.png|480px]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0078
|
| [[File:cnf_ml_cnn_denoised_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track.png|480px]]
|}

'''Train on multiple and test on 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Validation loss
! Inference Time
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0010
| 0.0007
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_seg5.png|380px]]

[[File:0.noise_histogram_mult23.png|480px]]
[[File:1.noise_histogram_mult23.png|480px]]
[[File:2.noise_histogram_mult23.png|480px]]
[[File:3.noise_histogram_mult23.png|480px]]
[[File:4.noise_histogram_mult23.png|480px]]
[[File:5.noise_histogram_mult23.png|480px]]

'''Train on new 2-3 tracks and test on multitrack V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.003
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new_multitrack.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_multitrack_new_seg5.png|380px]]

[[File:0.noise_histogram_multitrack_new.png|480px]]
[[File:1.noise_histogram_multitrack_new.png|480px]]
[[File:2.noise_histogram_multitrack_new.png|480px]]
[[File:3.noise_histogram_multitrack_new.png|480px]]
[[File:4.noise_histogram_multitrack_new.png|480px]]
[[File:5.noise_histogram_multitrack_new.png|480px]]

'''Train on new 2-3 tracks and test on old 2-3 tracks V2'''

{| class="wikitable
! Epochs
! Training loss
! Validation loss
! Training Time
! Inference Time
! Train/Val Loss Plot
! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 20
| 0.0043
| 0.0045
| 517.2
| 0.0003
| [[File:cnn_ae_train_val_mult_loss_more_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_V2_new.png|480px]]
|}

[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg0.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg1.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg2.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg3.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg4.png|380px]]
[[File:cnf_ml_cnn_v2_hits_histogram_mult23_new_seg5.png|380px]]

[[File:0.noise_histogram_mult23_new.png|480px]]
[[File:1.noise_histogram_mult23_new.png|480px]]
[[File:2.noise_histogram_mult23_new.png|480px]]
[[File:3.noise_histogram_mult23_new.png|480px]]
[[File:4.noise_histogram_mult23_new.png|480px]]
[[File:5.noise_histogram_mult23_new.png|480px]]

'''Train on multiple and test on 2-3 tracks with more kernels'''

[[File:cnn_ae_model_more_kernels.png]]

{| class="wikitable
! Epochs
! Validation loss
! Inference Time

! Example Case
! Valid Hits Histogram
! Noise Added Histogram
|-
| 50
| 0.0075
| 0.0008

| [[File:cnf_ml_cnn_denoised_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_histogram_10_train_test_mult2_3_track_more.png|480px]]
| [[File:cnf_ml_cnn_denoised_noise_histogram_10_train_test_mult2_3_track_more.png|480px]]
|}

=== Examples ===

{| class="wikitable
! Noisy
! Correct
! MLP
! CNN
! CNN V2
! CNN V2 new
|-
| [[File:ae_noisy_109.png|360px]]
| [[File:ae_correct_109.png|360px]]
| [[File:ae_mlp_denoised_109.png|360px]]
| [[File:ae_cnn_denoised_109.png|360px]]
|-
| [[File:ae_noisy_1019.png|360px]]
| [[File:ae_correct_1019.png|360px]]
| [[File:ae_mlp_denoised_1019.png|360px]]
| [[File:ae_cnn_denoised_1019.png|360px]]
|-
| [[File:ae_noisy_312.png|360px]]
| [[File:ae_correct_312.png|360px]]
| [[File:ae_mlp_denoised_312.png|360px]]
| [[File:ae_cnn_denoised_312.png|360px]]
|-
| [[File:ae_noisy_666.png|360px]]
| [[File:ae_correct_666.png|360px]]
| [[File:ae_mlp_denoised_666.png|360px]]
| [[File:ae_cnn_denoised_666.png|360px]]
|-
| [[File:ae_noisy_1.png|360px]]
| [[File:ae_correct_1.png|360px]]
| [[File:ae_mlp_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1.png|360px]]
| [[File:ae_cnn_denoised_1_V2.png|360px]]
| [[File:ae_cnn_denoised_1_V2_new.png|360px]]
|-
| [[File:ae_noisy_2.png|360px]]
| [[File:ae_correct_2.png|360px]]
| [[File:ae_mlp_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2.png|360px]]
| [[File:ae_cnn_denoised_2_V2.png|360px]]
| [[File:ae_cnn_denoised_2_V2_new.png|360px]]
|-
| [[File:ae_noisy_3.png|360px]]
| [[File:ae_correct_3.png|360px]]
| [[File:ae_mlp_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3.png|360px]]
| [[File:ae_cnn_denoised_3_V2.png|360px]]
| [[File:ae_cnn_denoised_3_V2_new.png|360px]]
|-
| [[File:ae_noisy_1m.png|360px]]
| [[File:ae_correct_1m.png|360px]]
| [[File:ae_mlp_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m.png|360px]]
| [[File:ae_cnn_denoised_1m_V2.png|360px]]
| [[File:ae_cnn_denoised_1m_V2_new.png|360px]]
|-
| [[File:ae_noisy_2m.png|360px]]
| [[File:ae_correct_2m.png|360px]]
| [[File:ae_mlp_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m.png|360px]]
| [[File:ae_cnn_denoised_2m_V2.png|360px]]
| [[File:ae_cnn_denoised_2m_V2_new.png|360px]]
|-
| [[File:ae_noisy_3m.png|360px]]
| [[File:ae_correct_3m.png|360px]]
| [[File:ae_mlp_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m.png|360px]]
| [[File:ae_cnn_denoised_3m_V2.png|360px]]
| [[File:ae_cnn_denoised_3m_V2_new.png|360px]]
|-
| [[File:ae_noisy_4m.png|360px]]
| [[File:ae_correct_4m.png|360px]]
| [[File:ae_mlp_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m.png|360px]]
| [[File:ae_cnn_denoised_4m_V2.png|360px]]
| [[File:ae_cnn_denoised_4m_V2_new.png|360px]]
|}

=== Work Plan ===

* create a histogram comparing how many points in the reconstructed image match points in the ground truth
* create a histogram comparing how many points in reconstructed image are noise (are not in the original image)
* try network (trained with one track) with new data set that contains more than one track, and provide example plots for presentation
* create the efficiency and noise plots for the multi-track data sample reconstruction.
* train network with multi-track data and evaluate performance on multi-track testing sample.
* put on the web site architecture of the network and try with a little deeper architecture (4032->2016->1008->336 -> 224 -> 336 -> 1008 -> 2016 -> 4032)

=== Work Plan (November 17 ===

* Provide workflow for testing, given input image write out the truth and AI inferred image
* Write software to create an image from AI inferred image and Truth image to reconstruct better truth, by filling in missing pieces in AI produced image from truth image when the points are close
* Evaluate efficiency of reconstructing the truth image and noisiness of resulting image. Also make graph now much noise was removed from original image.
* Start working with Hall-C data. much bigger input and evaluate efficiency of cleaning noise
* at the first stage noise has to be added to the Hall-C data for training, it can start with one track sample, then 250 track sample.

= Resources =
https://userweb.jlab.org/~gavalian/Documents/NN_Vary_Width_Depth%20(Nguyen).pdf

https://userweb.jlab.org/~gavalian/ML/2020/AutoEncoder/

File:CNF 3D QUANTIZATION 24.png

2020-12-02T20:34:20Z

Angelos:

File:CNF 3D QUANTIZATION 36.png

2020-12-02T20:31:14Z

Angelos:

File:CNF 3D NO QUANTIZATION.png

2020-12-02T20:28:54Z

Angelos:

Main Page

2020-11-24T19:17:08Z

Angelos: /* Announcements */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Featured Video===
<html>
<div style="text-align:center;">
<iframe width="700" height="500" src="https://www.youtube-nocookie.com/embed/pnnZJgNs1kU" frameborder="0" allow="accelerometer; autoplay; clipboard-write; encrypted-media; gyroscope; picture-in-picture" allowfullscreen>
</iframe>
</div>
</html>

===Molecular Dynamics===
==== SARS-CoV-2 D614 (original strain) Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S5-D614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 (new strain) Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S6-G614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | DATA SCIENCE ROADMAP TO COMPTON FORM FACTORS OF QUARKS AND GLUONS
|-
|Chrisochoides' Real-Time Computing (CRTC) lab in collaboration with Dr. Charles Hyde (ODU), Dr. Simonetta Liuti (UVA), and Dr. David Heddle (CNU) will host a two-day mini-workshop on the roadmap to Compton Form Factors for efficient evaluations of Monte Carlo Simulations. For more information see: '''[[DATA_SCIENCE_ROADMAP_TO_COMPTON_FORM_FACTORS_OF_QUARKS_AND_GLUONS]]''' [[File:logo.png|250px| center]]
|}

{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
||The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]
|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

CNF:ML Data Denoise

2020-11-11T19:18:36Z

Angelos:

CNF:ML Data Denoise

2020-11-11T19:17:35Z

Angelos:

File:Cnf ml cnnv2 arch.png

2020-11-11T19:15:52Z

Angelos:

CNF:ML Data Denoise

2020-11-11T02:17:55Z

Angelos: /* Convolutional Networks */

File:Cnf ml cnn v2 cnn denoised.png

2020-11-11T02:14:50Z

Angelos:

File:Cnf ml cnn v2 hits histogram.png

2020-11-11T02:14:25Z

Angelos:

File:Cnf ml cnn v2 noise histogram.png

2020-11-11T02:14:21Z

Angelos:

File:Cnf ml cnn v2 train val loss.png

2020-11-11T02:14:15Z

Angelos:

Main Page

2020-11-05T21:37:31Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Dynamics===
==== SARS-CoV-2 D614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S5-D614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S6-G614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | DATA SCIENCE ROADMAP TO COMPTON FORM FACTORS OF QUARKS AND GLUONS
|-
|Chrisochoides' Real-Time Computing (CRTC) lab in collaboration with Dr. Charles Hyde (ODU), Dr. Simonetta Liuti (UVA), and Dr. David Heddle (CNU) will host a two-day mini-workshop on the roadmap to Compton Form Factors for efficient evaluations of Monte Carlo Simulations. For more information see: '''[[DATA_SCIENCE_ROADMAP_TO_COMPTON_FORM_FACTORS_OF_QUARKS_AND_GLUONS]]''' [[File:logo.png|250px| center]]
|}

{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
||The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]
|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-11-05T21:37:14Z

Angelos: /* Molecular Dynamics */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Dynamics===
==== SARS-CoV-2 D614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S5-D614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S6-G614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | DATA SCIENCE ROADMAP TO COMPTON FORM FACTORS OF QUARKS AND GLUONS
|-
|Chrisochoides' Real-Time Computing (CRTC) lab in collaboration with Dr. Charles Hyde (ODU), Dr. Simonetta Liuti (UVA), and Dr. David Heddle (CNU) will host a two-day mini-workshop on the roadmap to Compton Form Factors for efficient evaluations of Monte Carlo Simulations. For more information see: '''[[DATA_SCIENCE_ROADMAP_TO_COMPTON_FORM_FACTORS_OF_QUARKS_AND_GLUONS]]''' [[File:logo.png|250px| center]]
|}

{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
||The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]
|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-11-05T21:36:50Z

Angelos: /* Announcements */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Dynamics===
==== SARS-CoV-2 D614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S5-D614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 Spike Cleavage Site Interaction with Furin ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/S6-G614-w-furin.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | DATA SCIENCE ROADMAP TO COMPTON FORM FACTORS OF QUARKS AND GLUONS
|-
|Chrisochoides' Real-Time Computing (CRTC) lab in collaboration with Dr. Charles Hyde (ODU), Dr. Simonetta Liuti (UVA), and Dr. David Heddle (CNU) will host a two-day mini-workshop on the roadmap to Compton Form Factors for efficient evaluations of Monte Carlo Simulations. For more information see: '''[[DATA_SCIENCE_ROADMAP_TO_COMPTON_FORM_FACTORS_OF_QUARKS_AND_GLUONS]]''' [[File:logo.png|250px| center]]
|}

{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
||The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]
|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Events

2020-08-25T19:52:43Z

Angelos:

= Seminars =
== Optimistic Cloud & Edge Computing Outside Hardware Boundaries ==
'''Date:''' October 10, 2019

'''Presenter:''' Professor Dimitrios S. Nikolopoulos

'''Abstract:'''
To address scaling limitations of future hardware, computing systems turned to parallelism and distribution. Most of the software and applications in science and engineering, but also applications that we use in our daily lives are actually distributed programs with some components running on edge or IoT devices to serve clients, data collectors or actuators, and other components running on data centers to provide data analytics, simulation, or visualization. The disaggregation of computing services raises new challenges for system challenges. We explores two of these challenges in this talk and discuss some solutions. The first challenge is that many applications necessitate low latency and more analytical power at or near the data sources. We demonstrate a system called TAPAS, which is neural network architecture search exploration engine. TAPAS uses aggressive compression, approximation and learning techniques to avoid entirely the simulation process in exploring neural network architectures. It further uses learning methods to adapt immediately to unseen data sets. TAPAS runs on a single low-power GPU and can train over 1,000 networks per second. This makes TAPAS suitable for training machine learning models on edge devices with limited resources. The second challenge is the one of scaling the performance and energy-efficiency of the hardware used in the Cloud and the Edge beyond current boundaries. We explore a co-designed compiler/OS/firmware system for characterizing hardware operating boundaries and safely operating hardware outside those boundaries to gain performance at the expense of additional, yet infrequent errors and mitigating actions. We demonstrate that many applications are inherently resilient to extended hardware boundaries and indeed benefit substantially from boundary relaxation.
<li style="display: inline-block;">
[[File: dimitrios.jpg|frameless|left|200px]]
'''Bio''': Dimitrios Nikolopoulos is a Professor of Engineering and he was recently named the John W. Hancock Professor of Engineering by the Virginia Tech Board of Visitors. He received his bachelor’s degree, master’s degree, and Ph.D. from the University of Patras. He spent the past 10 years in Europe, most recently as a professor of high performance and distributed computing and director of the Institute on Electronics, Communications, and Information Technology at Queen's University Belfast. He brings to Virginia Tech a world-class record of scholarship, teaching, service, and outreach. Through fundamental scholarship on computer systems, Nikolopoulos has made contributions to the global computing systems research community. He has published 55 peer-reviewed journal articles and 122 peer-reviewed papers in highly regarded archival conference proceedings. Nikolopoulos has advised or co-advised 22 Ph.D. students through completion and continues to advise six Ph.D. students. He has also advised 16 postdoctoral research fellows. He is a Distinguished Member of the Association for Computing Machinery (ACM) and is a recipient of a Royal Society Wolfson Research Merit Award, a National Science Foundation CAREER Award, a Department of Energy CAREER Award, and an IBM Faculty Award.</li>

== Serious Games and Simulations: applications, challenges and future directions ==
'''Date:''' TBD, 2020

'''Presenter:''' Prof. Anastasia Angelopoulou

'''Abstract:''' Serious games and simulations have been steadily increasing their
use in many sectors of society, particularly in education, defense, science and health. Their main purpose is usually to educate or train the users. In this talk, I will present my work in the area of serious games and simulations for training. I will also discuss challenges in the serious games development and future directions to overcome them.

<li style="display: inline-block;">
[[File: Anastasia.jpg|frameless|left|150px]]
'''Short Bio:''' Anastasia Angelopoulou is an Assistant Professor in Simulation and Gaming at the TSYS School of Computer Science at Columbus State University (CSU). Prior to joining CSU, she was a postdoctoral associate at the Institute for Simulation and Training at University of Central Florida (2016-2018), where she obtained her MSc and PhD in Modeling and Simulation (2015). Her research interests lie in the areas of modeling and simulation and serious games and their applications in domains such as healthcare, military, energy, and education, among others. Her research work has been partially supported by the Office of Naval Research and the National Science Foundation (NSF).
</li>

== Black-Box Analysis ==
'''Date:''' TBD, 2020

'''Presenter:''' Dr. Daniele Panozzo

'''Abstract:''' The numerical solution of partial differential equations (PDE) is ubiquitous in computer graphics and engineering applications, ranging from the computation of UV maps and skinning weights, to the simulation of elastic deformations, fluids, and light scattering. Ideally, a PDE solver should be a “black box”: the user provides as input the domain boundary, boundary conditions, and the governing equations, and the code returns an evaluator that can compute the value of the solution at any point of the input domain. This is surprisingly far from being the case for all existing open-source or commercial software, despite the research efforts in this direction and the large academic and industrial interest. To a large extent, this is due to treating meshing and FEM basis construction as two disjoint problems.

I will present an integrated pipeline, considering meshing and element design as a single challenge, that makes the tradeoff between mesh quality and element complexity/cost local, instead of making an a priori decision for the whole pipeline. I will demonstrate that tackling the two problems jointly offers many advantages, and that a fully black-box meshing and analysis solution is already possible for heat transfer and elasticity problems.

<li style="display: inline-block;">
[[File: Daniele.jpg|frameless|left|150px]]
'''Short Bio:''' Dr. Daniele Panozzo is an Assistant Professor of Computer Science at the Courant Institute of Mathematical Sciences in New York University. Prior to joining NYU he was a postdoctoral researcher at ETH Zurich (2012-2015). Daniele earned his PhD in Computer Science from the University of Genova (2012) and his doctoral thesis received the EUROGRAPHICS Award for Best PhD Thesis (2013). He received the EUROGRAPHICS Young Researcher Award in 2015 and the NSF CAREER Award in 2017. Daniele is leading the development of libigl (https://github.com/libigl/libigl), an award-winning (EUROGRAPHICS Symposium of Geometry Processing Software Award, 2015) open-source geometry processing library that supports academic and industrial research and practice. Daniele is chairing the Graphics Replicability Stamp (http://www.replicabilitystamp.org), which is an initiative to promote reproducibility of research results and to allow scientists and practitioners to immediately beneﬁt from state-of-the-art research results. His research interests are in digital fabrication, geometry processing, architectural geometry, and discrete differential geometry.
</li>

== New Approaches to Energy-Efficient and Resilient HPC ==
'''Date''': Nov 12,2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Dimitrios S. Nikolopoulos

'''Abstract''': This talk explores new and unconventional directions towards improving the energy-efficiency of HPC systems. Taking a workload-driven approach, we explore micro-servers with programmable accelerators; non-volatile main memory; workload auto-scaling and structured approximate computing. Our research in these has achieved significant gains in energy-efficiency while meeting application-specific QoS targets. The talk also reflects on a number of UK and European efforts to create a new energy-efficient and disaggregated ICT ecosystem for data analytics.
<li style="display: inline-block;">
[[File: Nikolopoulos.jpg|frameless|left|100px]]
'''Bio''': Dimitrios S. Nikolopoulos is Professor in the School of EEECS, at Queen's University of Belfast and a Royal Society Wolfson Research Fellow. He holds the Chair in High Performance and Distributed Computing and directs the HPDC Research Cluster, a team of 20 academic and research staff. His research explores scalable computing systems for data-driven applications and new computing paradigms at the limits of performance, power and reliability. Dimitrios received the NSF CAREER Award, the DOE CAREER Award, and the IBM Faculty Award during an eight-year tenure in the United States. He has also been awarded the SFI-DEL Investigator Award, a Marie Curie Fellowship, a HiPEAC Fellowship, and seven Best Paper Awards including some from the leading IEEE and ACM conferences in HPC, such as SC, PPoPP, and IPDPS. His research has produced over 150 top-tier outputs and has received extensive (£10.6m as PI/£39.5m as CoI) and highly competitive research funding from the NSF, DOE, EPSRC, SFI, DEL, Royal Academy of Engineering, Royal Society, European Commission and private sector. Dimitrios is a Fellow of the British Computer Society, Senior Member of the IEEE and Senior Member of the ACM. He earned a PhD (2000) in Computer Engineering and Informatics from the University of Patras. </li>

== Flow Diverters to Cure Cerebral Aneurysms a Case Study - From Concept to Clinical ==
'''Date''': Nov. 6, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Lieber, Baruch Barry

'''Abstract''': en to fifteen million Americans are estimated to harbor intracranial aneurysms (abnormal bulges of blood vessels located in the brain) that can rupture and expel blood directly into the brain space outside of the arteries causing a stroke. A flow diverter, a refined tubular mesh-like device that is inserted through a small incision in the groin area (no need for open brain surgery) and navigated through a catheter into cerebral arteries to treat brain aneurysms is delivered into the artery carrying the aneurysm. The permeability of the device is optimized such that it significantly reduces the blood flow in the aneurysm, while keeping small side branches of the artery open to supply critical brain tissue. The biocompatible device elicits a healthy scar-response from the body that lines the inner metal surface of the device with biological tissue, thus restoring the diseased arterial segment to its normal state. Refinement in the design of such devices and prediction of their long term creative effect, which usually occurs over a period of months can be significantly helped by computer modeling and simulations of the flow alteration such devices impart to the aneurysm. The evolution of these devices will be discussed from conception to their current clinical use.
<li style="display: inline-block;">
[[File: LieberB.jpg|frameless|left|125px |'''Professor Lieber, Baruch Barry''' ]]
'''Bio:''' Barry Lieber attended Tel-Aviv University and received a B.Sc. in Mechanical Engineering in 1979. He then attended Georgia Tech and received M.Sc. in 1982 and a Ph.D. in 1985, both in Aerospace Engineering Ph.D. working with Dr. Don P. Giddens. Barry Lieber was a Postdoctoral Fellow from 1985-1987 at the Department of Mechanical Engineering at Georgia Tech and also completed a summer fellowship at Imperial College London in 1986. In 1987 Barry Lieber joined the faculty of the Department of Mechanical and Aerospace Engineering at the State University of New York at Buffalo as Assistant Professor. In 1993 he was promoted to the rank of Associate Professor with tenure and in 1998 was promoted to full professor. In 1994 he became Research Professor of Neurosurgery and in 1997 he became the Director of the Center for Bioengineering at the State University of New York at Buffalo, both position he held until his departure from the university in 2001 to Join the University of Miami as professor in the Department of Biomedical Engineering with a joined appointment in the Department of Radiology. In 2010 he joined the State University of New York at Stony Brook at the rank of professor in the department of Neurosurgery and also serves as program faculty in the department of Biomedical Engineering. Barry Lieber was elected as fellow of the American Institute for Medical and Biomedical Engineering in 1999. He was elected as fellow of the American Society of mechanical Engineers in 2005 and served as the Chairman of the Division of Bioengineering of the American Society of Mechanical Engineers in 2009. </li>

== Enhanced Surface Definition in Moving-Boundary Flow Simulation ==
'''Date''': July 31, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Marek Behr

'''Abstract''': Moving-boundary flow simulations are an important design and analysis tool in many areas of engineering, including civil and biomedical engineering, as well as production engineering [1]. While interface-capturing offers unmatched flexibility for complex free-surface motion, the interface-tracking approach is very attractive due to its better mass conservation properties at low resolution. We focus on interface-tracking moving-boundary flow simulations based on stabilized discretizations of Navier-Stokes equations, space-time formulations on moving grids, and mesh update mechanisms based on elasticity. However, we also develop techniques that promise to increase the fidelity of the interface-capturing methods.

In order to obtain accurate and smooth shape description of the free surface, as well as accurate flow approximation on coarse meshes, the approach of NURBS-enhanced finite elements (NEFEM) [2] is being applied to various aspects of free-surface flow computations. In NEFEM, certain parts of the boundary of the computational domain are represented using non-uniform rational B-splines (NURBS), therefore making it an effective technique to accurately treat curved boundaries, not only in terms of geometry representation, but also in terms of solution accuracy.

As a step in the direction of NEFEM, the benefits of a purely geometrical NURBS representation of the free-surface could already be shown [3]. The first results with a full NEFEM approach for the flow variables in the vicinity of the moving free surface have also been obtained. The applications include both production engineering, i.e., die swell in plastics processing simulation, and safety engineering, i.e., sloshing phenomena in fluid tanks subjected to external excitation.

Space-time approaches offer some not-yet-fully-exploited advantages when compared to standard discretizations (finite-difference in time and finite-element in space, using either method of Rothe or method of lines); among them, the potential to allow some degree of unstructured space-time meshing. A method for generating simplex space-time meshes is presented, allowing arbitrary temporal refinement in selected portions of space-time slabs. The method increases the flexibility of space-time discretizations, even in the absence of dedicated space-time mesh generation tools. The resulting tetrahedral (for 2D problems) and pentatope (for 3D problems) meshes are tested in the context of advection-diffusion equation, and are shown to provide reasonable solutions, while enabling varying time refinement in portions of the domain [4].

[1] S. Elgeti, M. Probst, C. Windeck, M. Behr, W. Michaeli, and C. Hopmann, "Numerical shape optimization as an approach to extrusion die design", Finite Elements in Analysis and Design, 61, 35–43 (2012).

[2] R. Sevilla, S. Fernandez-Mendez and A. Huerta, "NURBS-Enhanced Finite Element Method (NEFEM)", International Journal for Numerical Methods in Engineering, 76, 56–83 (2008).

[3] S. Elgeti, H. Sauerland, L. Pauli, and M. Behr, "On the Usage of NURBS as Interface Representation in Free-Surface Flows", International Journal for Numerical Methods in Fluids, 69, 73–87 (2012).

[4] M. Behr, "Simplex Space-Time Meshes in Finite Element Simulations", International Journal for Numerical Methods in Fluids, 57, 1421–1434, (2008).
<li style="display: inline-block;">
[[File: Visitor_Marek_Behr1.jpg|frameless|left|100px]]
'''Bio:''' Prof. Marek Behr obtained his Bachelor's and Ph.D. degrees in Aerospace Engineering and Mechanics form the University of Minnesota in Minneapolis. After faculty appointments at the University of Minnesota and at Rice University in Houston, he was appointed in 2004 as a Professor of Mechanical Engineering and holder of the Chair for Computational Analysis of Technical Systems at the RWTH Aachen University. Since 2006, he is the Scientific Director of the Aachen Institute for Advanced Study in Computational Engineering Science, focusing on inverse problems in engineering and funded in the framework of the Excellence Initiative in Germany. Behr advises or has advised over 40 doctoral students, and has published over 65 refereed journal articles and a similar number of conference publications and book chapters. Behr is one of the main developers of the stabilized space-time finite element formulation for deforming-domain flow problems, which has been recently extended to unstructured space-time meshes. He is a long-time expert on parallel computation and large-scale flow simulations and on numerical methods for non-Newtonian fluids. He is a member of several advisory and editorial boards of international journals, and the member of the executive council of the German Association for Computational Mechanics and of the general council of the International Association for Computational Mechanics. </li>

== Disrupting the power/performance/quality tradeoff through approximate and error-tolerant computing ==
'''Date''': June 25, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Christos Antonopoulos

'''Abstract''': A major obstacle in the path towards exascale computing is the necessity to improve the energy efficiency of systems by two orders of magnitude. Embedded computing also faces similar challenges, in an era when traditional techniques, such as DVFS and Vdd scaling, yield very limited additional returns. Heterogeneous platforms are popular due to their power efficiency. They usually consist of a host processor and a number of accelerators (typically GPUs). They may also integrate multiple cores or processors with inherently different characteristics, or even just configured differently. Additional energy gains can be achieved for certain classes of applications by approximating computations, or in a more aggressive setting even tolerating errors. These opportunities, however, have to be exploited in a careful, educated manner, otherwise they may introduce significant development overhead and may also result to catastrophic failures or uncontrolled degradation of the quality of results. Introducing and tolerating approximations and errors in a disciplined and effective way requires rethinking, redesigning and re-engineering all layers of the system stack, from programming models down to hardware. We will present our experiences from this endeavor in the context of two research projects: Centaurus (co-funded by GR an EU) and SCoRPiO (EU FET-Open). We will also discuss our perspective on the main obstacles preventing the wider adoption of approximate and error-aware computing and the necessary steps to be taken to that end.
<li style="display: inline-block;">
[[File: Antonopoulos.jpg|frameless|left|100px]]
'''Bio''': Christos D. Antonopoulos, is Assistant Professor at the Department of Electrical and Computer Engineering of the University of Thessaly in Volos, Greece. He earned his PhD (2004), MSc (2001) and Diploma (1998) from the Department of Computer Engineering and Informatics of the University of Patras, Greece. His research interests span the areas of system and applications software for high performance computing, emphasizing on monitoring and adaptivity with performance and power/performance/quality criteria. He is the author of more than 50 refereed technical papers, and has been awarded two best-paper awards. He has been actively involved in several research projects both in the EU and in USA. </li>

== Image-Based Mesh Generation and Volumetric Spline Modeling for Isogeometric Analysis ==
'''Date''': April 24, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Yongjie Jessica Zhang

'''Abstract''': With finite element methods and scanning technologies seeing increased use in many research areas, there is an emerging need for high-fidelity geometric modeling and mesh generation of spatially realistic domains. In this talk, I will highlight our research in three areas: image-based mesh generation for complicated domains, trivariate spline modeling for isogeometric analysis, as well as biomedical, material sciences and engineering applications. I will first present advances and challenges in image-based geometric modeling and meshing along with a comprehensive computational framework, which integrates image processing, geometric modeling, mesh generation and quality improvement with multi-scale analysis at molecular, cellular, tissue and organ scales. Different from other existing methods, the presented framework supports five unique features: high-fidelity meshing for heterogeneous domains with topology ambiguity resolved; multiscale geometric modeling for biomolecular complexes; automatic all-hexahedral mesh generation with sharp feature preservation; robust quality improvement for non-manifold meshes; and guaranteed-quality meshing. These unique capabilities enable accurate, stable, and efficient mechanics calculation for many biomedicine, materials science and engineering applications.

In the second part of this talk, I will show our latest research on volumetric spline parameterization, which contributes directly to the integration of design and analysis, the root idea of isogeometric analysis. For arbitrary topology objects, we first build a polycube whose topology is equivalent to the input geometry and it serves as the parametric domain for the following trivariate T-spline construction. Boolean operations and geometry skeleton can also be used to preserve surface features. A parametric mapping is then used to build a one-to-one correspondence between the input geometry and the polycube boundary. After that, we choose the deformed octree subdivision of the polycube as the initial T-mesh, and make it valid through pillowing, quality improvement, and applying templates or truncation mechanism couple with subdivision to handle extraordinary nodes. The parametric mapping method has been further extended to conformal solid T-spline construction with the input surface parameterization preserved and trimming curves handled.
<li style="display: inline-block;">
[[File: Jessica.jpg|frameless|left|120px]]
'''Bio''': Yongjie Jessica Zhang is an Associate Professor in Mechanical Engineering at Carnegie Mellon University with a courtesy appointment in Biomedical Engineering. She received her B.Eng. in Automotive Engineering, and M.Eng. in Engineering Mechanics, all from Tsinghua University, China, and M.Eng. in Aerospace Engineering and Engineering Mechanics, and Ph.D. in Computational Engineering and Sciences from the University of Texas at Austin. Her research interests include computational geometry, mesh generation, computer graphics, visualization, finite element method, isogeometric analysis and their application in computational biomedicine, material sciences and engineering. She has co-authored over 100 publications in peer-reviewed journals and conference proceedings. She is the recipient of Presidential Early Career Award for Scientists and Engineers, NSF CAREER Award, Office of Naval Research Young Investigator Award, USACM Gallagher Young Investigator Award, Clarence H. Adamson Career Faculty Fellow in Mechanical Engineering, George Tallman Ladd Research Award, and Donald L. & Rhonda Struminger Faculty Fellow. </li>

== AFLR Unstructured Meshing Research Activities at CFD Modeling and Simulation Research at the Center for Advanced Vehicular Systems ==
'''Date''': March 20, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor David Marcum

'''Abstract''': Mesh generation and associated geometry preparation are critical aspects of any computational field simulation (CFS) process. In particular the mesh used can have a significant impact on accuracy, effectiveness, and efficiency of the CFS solver. Further, typical users spend a considerable portion of their time for the overall effort on mesh and geometry issues. All of this is particularly critical for CFD applications. AFLR is an unstructured mesh generator designed with a focus on addressing these issues for complex geometries. It is widely used, readily available to Government and Academic users, and has been very successful with relevant problems. AFLR volume and surface meshing is also directly incorporated in several systems, including: DoD CREATE-MG Capstone, Lockheed Martin/DoD ACAD, Boeing MADCAP, MSU SolidMesh, and Altair HyperMesh. In this talk we will provide an overview of this technology, future directions, and plans for multi-tasking/parallel operation.
<li style="display: inline-block;">
[[File: Marcum David.jpg|frameless|left|125px]]
'''Bio''': Dr. Marcum is Professor of Mechanical Engineering at Mississippi State University (MSU) and Chief Scientist for CFD within the Center for Advanced Vehicular Systems (CAVS). He has 30 years of experience in development of CFD and unstructured grid technology. Before joining MSU in 1991, Dr. Marcum was a Scientist and Senior Engineer at McDonnell Douglas Research Laboratories and Boeing Commercial Airplane Company. He received his Ph.D. from Purdue University in 1985. Prior to that he was a Senior Engineer from 1978 through 1983 at TRW Ross Gear Division. At MSU, Dr. Marcum served as Thrust Leader and Director of the NSF ERC for Computational Field Simulation. As Director, he led the transition from graduated NSF ERC to its current form as the High Performance Computing Collaboratory (HPC²). Dr. Marcum also served as Deputy Director and Director of the SimCenter (an HPC² member center and currently merged within CAVS). He is currently Chief Scientist for CFD within CAVS (also an HPC² member center). As Chief Scientist for CFD, he is directly involved in the research activities of a team of multi-disciplinary researchers working on CFD related projects for DoD, DoE, NASA, NSF, and industry. Computational tools produced by these projects at MSU within the ERC, SimCenter and CAVS, and in particular Dr. Marcum’s AFLR unstructured mesh generator, are in use throughout aerospace, automotive and DoD organizations. Dr. Marcum is widely recognized for his contributions to unstructured grid technology and is currently Honorary Professor at University of Wales, Swansea, UK and a previous Invited Professor at INRIA, Paris-Rocquencourt, France. </li>

== Riemannian Optimization for Elastic Shape Analysis ==
'''Date''': January 23, 2015, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Kyle Gallivan

'''Abstract''': In elastic shape analysis, a representation of a shape is invariant to translation, scaling, rotation and reparameterization and important problems (such as computing the distance and geodesic between two curves, the mean of a set of curves, and other statistical analyses) require finding a best rotation and re-parameterization between two curves. In this talk, I focus on this key subproblem and study different tools for optimizations on the joint group of rotations and re-parameterizations. I will give a brief account of a novel Riemannian optimization approach and evaluate its use in computing the distance between two curves and classification using two public data sets. Experiments show significant advantages in computational time and reliability in performance compared to the current state-of-the-art method.
<li style="display: inline-block;">
[[File: Visitor_marcum.jpg|frameless|left|250px]]
'''Bio''': Kyle A. Gallivan is a Professor of Mathematics at Florida State University. Gallivan received the Ph.D. in Computer Science from the University of Illinois at Urbana-Champaign in 1983 under the direction of C. W. Gear. He worked on special purpose signal processors in the Government Aerospace Systems Division of Harris Corporation. He was a research computer scientist at the Center for Supercomputing Research and Development at the University of Illinois from 1985 until 1993 when he moved to the Department of Electrical and Computer Engineering. From 1997 to 2008 he was a member of the Department of Computer Science at Florida State University (FSU) and a member of the Computational Science and Engineering group becoming a full Professor in 1999. He became a Professor of Mathematics at FSU in 2008 and was selected the 2012 Pascal Professor for the Faculty of Sciences of the University of Leiden in the Netherlands. He has been a Visiting Professor at the Catholic University of Louvain in Belgium multiple times through a long-standing research collaboration with colleagues there.

Over the years Gallivan's research has included: design and analysis of high-performance numerical algorithms, pioneering work on block algorithms for numerical linear algebra, performance analysis of the experimental Cedar system, restructuring compilers, model reduction of large scale differential equations, and high-performance codes for application such as ocean circulation, circuit simulation and the codes in the Perfect Benchmark Suite. Gallivan's current main research concerns optimization algorithms on Riemannian manifolds and their use in applications such as shape analysis, statistics, and signal/image processing. </li>

== A Parallel Log Barrier for Mesh Quality Improvement and Updating ==
'''Date''': November 7, 2014, 10:30AM

'''Location''': E & CS Auditorium, First Floor

'''Presenter''': Professor Suzanne M. Shontz

'''Abstract''': There are numerous applications in science, engineering, and medicine which require high-quality meshes, i.e., discretizations of the geometry, for use in computational simulations. For example, meshes have been used to enable accurate prediction of the performance, reliability, and safety of solid propellant rockets. The movie industry in Hollywood typically employs dynamic meshes in order to animate characters in films. Large-scale applications often require meshes with millions to billions of elements that are generated and manipulated in parallel. The advent of supercomputers with hundreds to thousands of cores has made this possible.

The focus of my talk will be on parallel algorithms for mesh quality improvement and mesh untangling. Such algorithms are needed, for example, when a large-scale mesh deformation is applied and tangled and/or low quality meshes are the result. Prior efforts in these areas have focused on the development of parallel algorithms for mesh generation and local mesh quality improvement in which only one vertex is moved at a time. In contrast, we are concerned with the development of parallel global algorithms for mesh quality improvement and untangling in which all vertices are moved simultaneously. I will present our parallel log-barrier mesh quality improvement and untangling algorithms for distributed-memory machines. Our algorithms simultaneously move the mesh vertices in order to optimize a log-barrier objective function that was designed to improve the quality of the worst quality mesh elements. We employ an edge coloring-based algorithm for synchronizing unstructured communication among the processes executing the log-barrier mesh optimization algorithm. The main contribution of this work is a generic scheme for global mesh optimization. The algorithm shows greater strong scaling efficiency compared to an existing parallel mesh quality improvement technique. Portions of this talk represent joint work with Shankar Prasad Sastry, University of Utah, and Stephen Vavasis, University of Waterloo.
<li style="display: inline-block;">
[[File: Visitor_shontz.jpg|frameless|left|150px]]
'''Bio''':Suzanne M. Shontz is an Associate Professor in the Department of Electrical Engineering and Computer Science at the University of Kansas. She is also affiliated with the Graduate Program in Bioengineering and the Information and Telecommunication Technology Center. Prior to joining the University of Kansas in 2014, Suzanne was on the faculty at Mississippi State and Pennsylvania State Universities. She was also a postdoc at the University of Minnesota and earned her Ph.D. in Applied Mathematics from Cornell University.

Suzanne's research efforts focus centrally on parallel scientific computing, more specifically, the design and analysis of unstructured mesh, numerical optimization, model order reduction, and numerical linear algebra algorithms and their applications to medicine, images, electronic circuits, materials, and other applications. In 2012, she was awarded an NSF Presidential Early CAREER Award (i.e., NSF PECASE Award) by President Obama for her research in computational- and data-enabled science and engineering. Suzanne also received an NSF CAREER Award for her research on parallel dynamic meshing algorithms, theory, and software for simulation-assisted medical interventions in 2011 and a Summer Faculty Fellowship from the Office of Naval Research in 2009. She has chaired or co-chaired several top conferences in computational- and data-enabled science and engineering including the International Meshing Roundtable in 2010 and the NSF CyberBridges Workshop in 2012-2014 and has served on numerous program committees in the field. Suzanne is also an Associate Editor for the Book Series in Medicine by De Gruyter Open. </li>

= Workshops =

== Parallel Software Runtime System Workshop ==

'''Date''': 24-25 May 2017

'''Location''': NASA/LaRC & NIA

'''Participants''': Pete Beckman (ANL), Halim Amer (ANL), Dana P. Hammond (NASA LaRC), Nikos Chrisochoides (ODU), Andriy Kot (NCSA,UIUC), Fotis Drakopoulos (ODU), Thomas Kennedy (ODU), Christos Tsolakis (ODU), Kevin Garner (ODU), Polykarpos Thomadakis (ODU)

== Isotropic Advancing Front Local Reconnection Hands-On Workshop ==

'''Location''': March 20-21, 2015

'''Participants''': NASA/LaRC : Dr Bill Jones , Dr Mike Mark, Dr Dana Hamond, ODU: Nikos Chrisochoides, Fotis Drakopoulos (ODU), Thomas Kennedy (ODU), Christos Tsolakis (ODU), Kevin Garner (ODU), Polykarpos Thomadakis (ODU)

== HPC Middleware for Mesh Generation and High Order Geometry Approximation Workshop ==
Attendants: (NASA/LaRC) Dr Bill Jones , Dr Mike Mark, Dr Dana Hamond,(NIA) Boris Diskin ODU: Nikos Chrisochoides

: ''' Dr. Navamita Ray ''' 

:Los Alamos National Laboratory, Mathematics and Computer Science Division

:Los Alamos, New Mexico

:'''When''': March 25,2016, 10:30AM

:'''Where''': E & CS Auditorium, First Floor

:'''What''': Towards Scalable Framework for Geometry and Meshing in Scientific Computing

:'''Email''': [mailto:navamitaray@gmail.com navamitaray@gmail.com]

:'''ABSTRACT'''
:High fidelity computational modeling of complex, coupled physical phenomena occurring in several scientific fields require accurate resolution of intricate geometry features, generation of good quality unstructured meshes that minimize modeling errors, scalable interfaces to load/manipulate/traverse these meshes in memory and support I/O for check-pointing and in-situ visualization. While several applications tend to create custom HPC solutions to tackle the heterogeneous descriptions of physical models, such approaches lack in generality, interoperability and extensibility making it difficult to maintain scalability of the individual representations. In this talk, we introduce the component-based open-source '''SIGMA''' (Scalable Interfaces for Geometry and Mesh based Applications) toolkit, an effort to address these issues. We focus particularly on its array-based unstructured mesh representation component, Mesh Oriented datABase ('''MOAB''') that provides scalable interfaces to geometry, mesh and solvers to allow seamless integration to computational workflows.
:[[File: Navamita.jpg|frameless|left|120px]]Based on the three fundamental units consisting of 1) compact array-based memory management for mesh and field data,2) efficient mesh data structures for traverals and querying, and 3) scalable parallel communication algorithms for distributed meshes, MOAB supports various advanced algorithms such as I/O, in-memory mesh modification and refinement, multi-mesh projections, high-order boundary reconstruction, etc. We discuss some of these advanced algorithms and their applications.

:'''Bio''': Dr. Navamita Ray is a postdoctoral appointee and part of the SIGMA team at Mathematics and Computer Science Division at Argonne National Laboratory, Argonne, IL. She has been involved in research on flexible mesh data structures for mesh adaptivity as well as high-fidelity discrete boundary representation. Dr. Ray holds a Ph.D. in Applied Mathematics from the Stony Brook University, where she did graduate work on high-order surface reconstruction and its applications to surface integrals and remeshing. </li>

: '''Dr. Xiangmin (Jim) Jiao '''
:Associate Professor and AMS Ph.D. Program Director, Department of Applied Mathematics and Statistics and Institute for Advanced Computational Science

:Stony Brook University

:'''When''': March 3,2016, 10:30AM

:'''Where''': E & CS Auditorium, First Floor

:'''What''': Robust Adaptive High-Order Geometric and Numerical Methods Based on Weighted Least Squares

:'''Email''': [mailto:xiangmin.jiao@stonybrook.edu xiangmin.jiao@stonybrook.edu]:

:'''Homepage''': http://www.ams.sunysb.edu/~jiao

:'''ABSTRACT'''

:Numerical solutions of partial differential equations (PDEs) are important for modeling and simulations in many scientific and engineering applications. Their solutions over complex geometries pose significant challenges in efficient surface and volume mesh generation and robust numerical discretizations. In this talk, we present our recent work in tackling these challenges from two aspects. First, we will present accurate and robust high-order geometric algorithms on discrete surface, to support high-order surface reconstruction, surface mesh generation and adaptation, and computation of differential geometric operators, without the need to access the CAD models. Secondly, we present some new numerical discretization techniques, including a generalized finite element method based on adaptive extended stencils,and a novel essentially nonoscillatory scheme for hyperbolic conservation laws on unstructured meshes. These new discretizations are more tolerant of mesh quality and allow accurate, stable and efficient computations even on meshes with poorly shaped elements. Based on a unified theoretical framework of weighted least squares, these techniques can significantly simplify the mesh generation processes, especially on supercomputers, and also enable more efficient and robust numerical computations. We will present the theoretical foundation of our methods and demonstrate their applications for mesh generation and numerical solutions of PDEs.

<li style="display: inline-block;">
:[[File: Collaborator Jiao.jpg|frameless|left|100px]]
:'''Bio''': Dr. Xiangmin (Jim) Jiao is an Associated Professor in Applied Mathematics and Computer Science, and also a core faculty member of the Institute for Advanced Computational Science at Stony Brook University. He received his Ph.D. in Computer Science in 2001 from University of Illinois at Urbana-Champaign (UIUC). He was a Research Scientist at the Center for Simulation of Advanced Rockets (CSAR) at UIUC between 2001 and 2005, and then an Assistant Professor in College of Computing at Georgia Institute of Technology between 2005 and 2007. His research interests focus on high-performance geometric and numerical computing, including applied computational and differential geometry, generalized finite difference and finite element methods, multigrid and iterative methods for sparse linear systems, multiphysics coupling, and problem solving environments, with applications in computational fluid dynamics, structural mechanics, biomedical engineering, climate modeling, etc. </li>

== CNF Imaging Workshop ==

:'''When''': August 2019

:'''Where''': tbd

:'''More Information''': [[CNF_Imaging_Workshop | CNF Imaging Workshop ]]

= Outreach =

== Surgical Planning Lab ==
''' When ''' : April 8 & 9 , 2016
''' Where ''' : Brigham and Women's Hospital & Harvard Medical School, Boston

Posters presented in 25th anniversary of SPL :

Fotis Drakopoulos and Nikos Chrisochoides : [http://www.cs.odu.edu/crtc/papers/SPL25/Chrisochoides_CBC3D.pdf Lattice-Based Multi-Tissue Mesh Generation for Biomedical Applications]

Fotis Drakopoulos and Nikos Chrisochoides : [http://www.cs.odu.edu/crtc/papers/SPL25/Chrisochoides_NRR.pdf Deformable Registration of Pre-Op MRI with iMRI for Brain Tumor Resection: Progress Report]

Nikos Chrisochoides, Andrey Chernikov and Christos Tsolakis : [http://www.cs.odu.edu/crtc/papers/SPL25/Chrisochoides_Telescopic.pdf Extreme Scale Mesh Generation for Big-Data Medical Images]

Internal:GPCD

2020-07-31T17:55:05Z

Angelos: /* Papers */

== Introduction ==

=== Protein Cavities ===
Cavities are structural features of proteins and they are often important for drug discovery and design. There are different kinds of cavities, including pockets, voids, tunnels, and channels. The types of cavities are listed in more detail in figure 1.

[[File:proteinCavityTypes.jpg|center|frame| Figure 1: The basic types of protein cavities [1]]]

# '''Void''': A cavity that looks like internal "hole" in a protein with no connection to the surface.
# '''Cleft''': A pocket that looks like a "dent" in the surface of a protein. Can also be found as part of a tunnel or channel.
# '''Invagination''': A pocket that looks like a "cove". It has a small opening on the surface of a protein and leads to a larger interior. Can also be found as part of a tunnel or after a cleft.
# '''Tunnel''': A pocket that has an opening on the surface of a protein and leads to somewhere in the interior. Has no opening on the other side.
# '''Channel''': A cavity that has an opening on one side of the surface of a protein and leads to another opening in another location on the surface of the protein.

Of course, the types of cavities listed in figure 1 are only the basic ones. Combinations of these are also possible as shown in figure 2.

[[File:proteinCavityAdvancedTypes.jpg|center|frame| Figure 2: Combination types of protein cavities [1]]]

=== Protein Cavity Detection Methods ===
There are several methods for detecting protein cavities. There are sphere based methods, grid based methods, grid-and-sphere based methods, surface based methods, grid-and-surface based methods, time-varying methods, GPU based methods, consensus based methods, and finally tessellation based methods. These algorithms are shown in more detail in figure 3. Tessellation based methods mainly rely on Voronoi tessellation. Some methods rely on weighted Voronoi tessellation because of its ability to more accurately represent the volume of atoms. The Voronoi tessellation in these methods is computed from a set of points representing the atomic centers of the underlying protein.

[[File:proteinCavityDetectionAlgorithms.jpg|center|frame| Figure 3: A hierarchy of protein cavity detection algorithms [1]]]

== Literature Review ==

=== MOLE ===

==== Method ====
The MOLE algorithm is specifically designed to identify tunnels and channels. Finding superficial cavities is a secondary concern and does so as part of the surface approximation step. MOLE utilizes the surface obtained by the refinement of the convex hull to find superficial cavities such as clefts and invaginations. In essence, the approximation of a more accurate surface and the detection of these superficial cavities are closely related. As tetrahedra are removed from the convex hull to get a more accurate surface approximation, a parameter called ''interior threshold'' controls how many superficial cavities are found. For example, if the number of tetrahedra removed from a specific part of the molecule exceeds this threshold, the algorithm marks these tetrahedra as part of a superficial cavity.

[[File:moleVoronoiTunnelPath.jpg|center|frame| Figure 4: Shows in 2D the Voronoi diagram (thin lines) and the convex hull (dotted line) of molecule represented by atoms (circles). Each Voronoi edge is evaluated by the cost function (numbers). The thick line represents the optimal path found by the Dijkstra's algorithm from the given starting point (black circle), i.e., centerline of found tunnel. [1]]]

For the more complicated types of cavities, tunnels and channels, MOLE uses the Voronoi diagram to identify them. Start and end points are defined as the centers of the deepest tetrahedra in each cavity. The depth of the tetrahedron is defined as the number of Voronoi edges from the closest boundary tetrahedron (on the surface). Tunnels/channels are computed as the shortest paths between all pairs of start and end points in the same cavity diagram.

[[File:MoleDijkstraWeightFunction.png|center|frame| Figure 5: Shows the weight function used to estimate the weight of each Voronoi edge for Dijkstra's algorihtm. l(e) is the length of the edge, d(e) is the distance of the edge to the closest atom Van der Waals sphere and ϵ is a small number to avoid division by zero [1]]]

==== Complexity ====
* Worst-case: O(N^2 log N), where N is the number of atoms in the molecule.
* Average: O(M log M), where M is the number of vertices in the Voronoi diagram (in the worst case, M = N^2, but in most cases M = O(N)). As a result of the use of the incremental algorithm and Hilbert curve ordering, the complexity of calculating the Delaunay triangulation of most molecular structures is O(N log N). Finally, the complexity of all the remaining steps of the algorithm is at most O(M log M).

=== MolAxis ===

==== Method ====
The MolAxis algorithm relies on Voronoi and Dijkstra's algorithm to find channels, similarly to MOLE. The algorithm consists of the following high-level steps:

# Approximate the molecule using a collection of '''fixed-size spheres'''. The number of balls needed depends on the required quality of the approximation. In the extreme case, where the user is interested in a high resolution, each atom is approximated using a single sphere.
# Compute the '''Voronoi diagram''' of the centers of the fixed-sized spheres. The diagram is intersected with a user-defined bounding sphere. This defines the area of interest.
# Discard the parts of the Voronoi diagram that intersect the molecule.
# Define a '''graph''', using the Voronoi vertices and edges. A weight is given to each edge, which depends on its length and on its distance from the surface of the molecule.
# Define a '''source point (root vertex)'''. The user can either supply a source point or allow MolAxis to automatically choose the source point to be the '''center of the largest cavity in the molecule''' (basically detect channels inside-out).
# Apply a shortest path optimization algorithm ('''Dijktra's algorithm''') on the graph, computing a tree called the "corridor tree".
# The desired channels are paths in the corridor tree. Duplicate channels are avoided by using the split distance parameter. The split distance bounds the distance to the least common ancestor of the channel and all previously reported channels.

[[File:MolAxisVoronoi.png|center|frame| Figure 6: Shows an example of fixed-size circles (light-blue) approximating varying-size circles and the pathway diagram of their centers. The discarded portion of the Voronoi diagram of their centers is depicted using dotted lines [2]. This figure also helps illustrate the usefulness of the Voronoi tessellation in computing pathways.]]

== Protein Energetics ==
=== Metrics ===
# '''Hydrophobicity''': Physical property of a molecule that is seemingly repelled from a mass of water. Hydrophilic molecules are attracted to water. Hydrophobicity plays a major role in protein folding. Hydrophobic molecules also tend to be nonpolar and prefer other neutral molecules and nonpolar solvents. Hydrophobicity is the most widely used metric.
# '''Hydrogen Bonds''': Weak bonds between two molecules resulting from an electrostatic attraction between a proton in one molecule and an electronegative atom in the other. We often see these bonds between oxygen and hydrogen.
# '''Van der Waals Interactions''': Also known as Van der Waals forces, they are forces driven by induced electrical interactions between two or more atoms or molecules that are very close to each other. Van der Waals interaction is the weakest of all intermolecular attractions between molecules. These forces are distance-dependent.
# '''Salt Bridges''': Non-covalent (no sharing of eletrons; electromagnetic) interaction between two ionized sites. It has two components: a hydrogen bond and an electrostatic interaction.
# '''Disulfide Bridges''': Strong bonds that can form between two cysteines (an amino acid). The strength of disulfide bridges helps stabilize a protein. Disulfide bridges are especially common in proteins that are secreted from cells.
# '''Charge''': Plays a role in strength of bonds. A higher charge means that an atom will have stronger forces in an electromagnetic field.
# '''Polarity''': Separation of electric charge leading to a molecule or its chemical groups having an electric dipole moment, with a negatively charged end and a positively charged end. Indicates the existence of polar bonds, which are covalent (sharing of electrons) and are quite strong. About [https://chem.libretexts.org/Bookshelves/Inorganic_Chemistry/Book%3A_Introduction_to_Inorganic_Chemistry/01%3A_Review_of_Chemical_Bonding/1.04%3A_Bond_Polarity_and_Bond_Strength bond polarity].

== Possible Methods that need to be Explored ==

=== 1. PODM Tessellation Using Van der Waals Surface Image (Angelos) ===

The first method that could be used for detecting protein cavities in parallel involves using PODM to create a tessellation of an image representation of the Van der Waals surface of a protein. The Van der Waals surface is an accurate representation of the surface of a protein and represents each atom as a sphere of a radius corresponding to its Van der Waals radius.

The steps for this method are outlined below:

# '''Compute the Van der Waals surface (Angelos)''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface (Angelos)''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again necessary.
# '''Compute Delaunay tessellation using PODM''': Once we have the image of the Van der Waals surface, we use PODM to compute the Delaunay tessellation in parallel. A low enough delta should be used to ensure that the surface is as accurately captured as possible.
# '''Compute Voronoi tessellation using PODM''': The next step is to compute the Voronoi tessellation as the dual of the Delaunay tessellation in parallel. The Voronoi tessellation is useful for quickly identifying tunnels using Dijkstra's algorithm.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=400px>
File:6vxxVanDerWaalsSurface.png | thumb | Figure 7: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).

File:COVID-19-Spike-Glycoprotein-6vxx,d=5.png | thumb | Figure 8: A tessellation based on the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). A highly-accurate surface tessellation is produced.
</gallery>

=== 2. PDT Tessellation Using Point Cloud of Atoms (Spiros) ===

The second method involves giving PDT/PODM a point cloud of atoms and creating a tessellation from that. This would require that the Delaunay triangulation is generated and a method for computing an accurate surface for the point cloud is employed. Computing this surface is one of the most challenging steps of this method. A method that could be employed is α-shapes, but this has the disadvantage of not guaranteeing connectivity between tetrahedra. Another method probably needs to be used that can guarantee connectivity between the tetrahedra (such as the one from MOLE 2.5).

The steps for this method are outlined below:

# '''Read PDB file and make point cloud of atoms (Angelos)''': The first step would be to read an input PDB file for a protein and create a point cloud of atoms. This involves either writing a basic PDB file reader or finding one we can use. However, writing a basic one is better because it skips the overhead of other full-featured readers as we only care about the atomic centers and potentially the type of atoms.
# '''Compute Delaunay triangulation using PDT/PODM''': The second step is to compute the Delaunay triangulation from the point cloud in parallel. This will produce an inaccurate surface that needs to be refined.
# '''Compute accurate surface (Spiros)''': The Delaunay triangulation on a point cloud of atoms produces a surface that does not reflect the actual surface of the protein. As a result, an accurate surface needs to computed by refining the triangulation. This would involve removing tetrahedra until an accurate surface is found. This should be done in parallel.
## '''MOLE method''':
### The molecular surface is approximated by iterative removal of boundary tetrahedra from the outermost layers (i.e., tetrahedra found at the interface between the molecule and the external environment).
### Boundary tetrahedra produced by the triangulation are removed in this step if they are sufficiently large to fit a sphere with a given probe radius
#### Note: A tetrahedron T fits a sphere S with probe radius r if the center C of sphere S can be placed inside the tetrahedron and the distance to all vertices of T is greater than or equal to the sum of r, with the Van der Waals radius of an atom corresponding to the given vertex.
##### Explanation: Given a boundary tetrahedron, if a Sphere S with center = Incenter C, and radius = probe radius
### Next, tetrahedra that are too small to fit a sphere with interior radius are removed.
#### Explanation: Given a non-boundary tetrahedron, if the radius of its inscribed sphere is smaller than (probe?) radius, then the tetrahedron needs to be removed.
### Remaining tetrahedra form one or more connected components. We call the components that contain at least one tetrahedron on the molecular surface cavity diagrams.
### It should be noted here that the cavity diagram is a purely geometrical concept to help identify regions of space (volume) that can contain tunnels and only very loosely corresponds to the cavities shown in Figure 1B).
# '''Compute Voronoi tesselation''': Once the accurate surface is determined the Voronoi tessellation must be computed. This step could be done before the accurate surface is computed, but we would also have to refine the Voronoi if that is to happen.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=300px>

File:6vxxPointCloud.jpeg|thumb| Figure 9: The atomic point cloud of the COVID-19 spike glycoprotein (6vxx). There are 23,694 atoms (points) in this image.

File:6vxxUnrefinedDelaunay.jpeg|thumb| Figure 10: An unrefined Delaunay triangulation of the COVID-19 spike glycoprotein (6vxx). Generated using the Delaunay3D filter in ParaView.

File:6vxxAlphaShapesDelaunay.jpeg|thumb| Figure 11: A Delaunay triangulation using α-shapes with α=3 of the COVID-19 spike glycoprotein (6vxx). Using α-shapes we can get a better approximation of the surface. Generated using the Delaunay3D filter in ParaView.
</gallery>

=== 3. Surface-Grid-based Using Van der Waals Surface Volumetric Image ===

The third method involves writing a new software for doing protein cavity detection. This software would use a volumetric image (e.g. of type .mha) representing the Van der Waals surface of a protein. Uses perhaps the most faithful representation of the Van der Waals surface, which is the ground truth for a molecular surface. As long as the Van der Waal surface generation algorithm is fast, this method has the potential of being both accurate and fast as it eliminates the overhead of tessellation. Good parallelization is important for this method because there are many more voxels than there are tetrahedra in a tessellation-based method. Because this method operates on voxels, it is a candidate for future GPU parallelization. This method is also a strong candidate for deep learning, much more than the tessellation based methods, because its output can easily be used for training a deep learning model in the future.

The steps for this method are outline below:

# '''Compute the Van der Waals surface''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again likely necessary in the future.
# '''Identify cavities in parallel''': Using the image voxels, the cavities now need to be identified in parallel (potentially using GPU parallelization in the future). It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as volumetric image''': The last step is to output the detected cavities as a volumetric important. It is important to only include the identified cavities in this image. This way, researchers can overlay the normal protein structure with the volume of the cavities and easily see where they are.

<gallery mode="packed" heights=250px>
File:vanDerWaalsSurfaceImage2.jpeg|center|thumb|800px| Figure 12: Volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). There are 15.5 million voxels in this volume.

File:ProteinCavityDetectionSurfaceMethodGridExample.png|center|thumb|800px| Figure 13: Shows an example grid for a slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). The grey grid represents the scalar field whereas the green grid represents possible partitioning of the grid for multithreading. Finer or coarser partitioning may be used depending on the degree of parallelism.

File:VanDerWaalsSurfaceVolumeAnim.gif|center|frame| Figure 14: An animation of a thicker slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). Potential cavities and voids are clearly visible.
</gallery>

== Challenges ==

There are several challenges that need to be addressed if we are to have a good parallel solution for detecting protein cavities.

=== 1. Expensive Pre-Processing ===

The parallel methods involving the Van der Waals surface, especially the one with the image input have expensive pre-processing. For large molecules, getting this surface can be slow because existing implementations are sequential and written in Python. We need to come up with a way to quickly generate this surface and its image representation. This pre-processing does not have to be done if we operated on the point cloud using method 2 described previously, leading to a major speedup in overall performance.

[[File:6vxxVanDerWaalsSurface.png|center|thumb|800px| Figure 15: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).]]

=== 2. Massive Parallelism (Stretch Goal) ===

The solution we will implement may need to support massive parallelism, ideally using MPI. The reason for this is that molecular dynamics simulations can potentially involve a massive number of atoms. If we consider the 200 million coronavirus model, then a simulation containing 5000 frames, would involve a total of 1 trillion atoms. If we are to detect the cavities in every frame in real-time, a massively parallel solution is necessary. There is no doubt about why 250,000 CPU cores were required for the COVID-19 simulation.

<gallery mode="packed" heights=400px>
File:molecularDynamicsSimulation1.gif|frame| Figure 16: An animation produced by a molecular dynamics simulation.

File:molecularDynamicsSimulation2.gif|frame| Figure 17: An animation produced by the COVID-19 molecular dynamics simulation.
</gallery>

== Candidate Method: Protein Cavity Identification using Triangulation, unsigned Distance, and Clustering (PCITDC) ==
The PCITDC method employs Delaunay triangulation to generate a coarse and fine surface of a protein (the fine surface is enclosed within the coarse surface). Then, the unsigned distance of the fine surface from the coarse surface is computed. This is done by iterating through every vertex (atom) in the fine surface and computing its distance from the closest point on the coarse surface. Finally, three-level clustering is employed to cluster atoms based on their distance values.

The algorithm contains the following high-level steps:
=== (Spiros) Compute Inaccurate and Accurate surfaces ===
# Solvent Excluded Surface ('''SES''') image
## Create a space filling image with spheres for each atom where their radius is: Van der Waals radius + probe radius [DONE]
### Note: Shared memory parallelism possibly be employed [TODO]
## Perform Euclidean Distance Transform on the space filling image [DONE]
### Note: Shared memory parallelism can be employed [DONE]
## Extract Solvent excluded surface from the EDT result by setting: [DONE]
### All pixel values < probe-radius = 0
### All pixel values >= probe-radius = 1
### Note: Shared memory parallelism can be employed [TODO]
# Inaccurate and Accurate Surface Candidates:
## (Accurate) Solvent Excluded Surface ('''SES''') [DONE]
### Computed using:
#### PODM3D
### Advantages:
#### Can control the number of elements on the surface
#### the edge lengths bounds are relatively small which helps in the clustering.
### Disadvantages:
#### Doesn't have a direct relation to the molecule
##### But can be fixed using a KD-tree
#### Surface may need to be smoothed
### Note: Shared memory parallelism can be employed [DONE]
## (Inaccurate) Convex Hull Surface ('''CHS''') [DONE]
### Computed using:
#### CGAL algorithm over the SES produced by PODM3D
### Advantages:
#### It's easy and fast to calculate
#### It's well defined mathematically
### Disadvantages:
#### The depth values need to be scaled properly in the Unsigned Distance step.
### Note: Shared memory parallelism can be employed [TODO]
## (Inaccurate) Course Solvent Excluded Surface ('''CSES''') [TODO]
### Computed using:
#### a bigger SES surface using a bigger probe-radius
### Advantages:
#### The depth values will probably be appropriate for the Unsigned Distance step.
### Disadvantages:
#### A recomputation of Spacing filling, EDT, values setting, and PODM3D steps are required
#### It's not well defined mathematically since it's an approximation of an approximation
### Note: Shared memory parallelism can be employed [TODO]

=== (Angelos/Spiros) Compute unsigned distance of the accurate from the inaccurate surface ===
# Algorithm:
## We iterate through every vertex (atom) in the accurate surface and compute its distance from the closest point on the inaccurate surface. This distance indicates the ''depth'' of a specific vertex. We apply Logarithmic scaling.
# Computed using:
## (Angelos) Using vtkStaticCellLocator (octree) [DONE]
### Note: Shared memory parallelism can be used [TODO]
## (Angelos/Spiros) Using vtkSignedDistance (signed distance) [TODO]
## (Spiros) Using path tracing in Delaunay triangulation [TODO]
### Algorithmic idea
#### Find the central cell of the triangulation and then traverse cells using visibility walk and jump and walk techniques in order to minimize the distance with the given point.
#### Once you reach the boundary, you will have a set of possible cells which include the minimum point, and afterward, you compute it.
### Note: Shared memory parallelism can be employed [TODO]

=== (Angelos/Spiros) Compute atom clusters based on unsigned distance values: ===
# (Angelos) A modified version of DBSCAN that finds neighbors based on not only their density but also their unsigned distance values.
## Variations:
### (Angelos) One-level clustering [DONE]
### (Angelos) Multi-level clustering to capture all the types of cavities
#### E.g. Three-level clustering. The first level consists of large surface cavities. The second level consists of small surface cavities. The third level consists of deep cavities (i.e. channels). [DONE]
## Can be improved by:
### Implementation-wise:
#### (Spiros) Using mean-cluster-distance value instead of first-cluster-point-distance value. [DONE]
### Parameters Tuning:
#### (Spiros/Angelos) Utilizing min/mean/median/max edge lengths to define '''eps / search-radius''' [TODO]
#### (Spiros) Utilizing distibution statistics of distances to define '''minimum points per cluster'''. [TODO]
#### (Angelos) Identify distance areas, e.g. (0-40), where cavities do not exist. [TODO]
## Note: parallelism can be employed, see [https://link.springer.com/article/10.1023/A:1009884809343 A Fast Parallel Clustering Algorithm for Large Spatial Databases PDBSCAN] [TODO]
# (Angelos/Spiros) [https://pdfs.semanticscholar.org/f730/194462fb6bfcd0b90d7f22a960ed58d77199.pdf A Novel Spatial Clustering Algorithm Based on Delaunay Triangulation] [TODO]
## Note: parallelism can be explored [TODO]
# (Angelos) [https://dl.acm.org/doi/abs/10.1145/304181.304187 OPTICS: ordering points to identify the clustering structure] [TODO]

=== (Angelos) Classify clusters based on the type of cavity they are: ===
# Algorithms:
## Unsigned distance based classification using mean distance cluster value [TODO]
### Note: Shared memory parallelism can be used [TODO]
## Analyze the shape of clusters to identify what type of cavity they are. [TODO]
### This can potentially be done using machine learning.
### Note: Shared memory parallelism can be used [TODO]

=== Output: ===
# Labeled fine surface [DONE]
# Labeled protein [TODO]
## Note: a Paraview plugin may be developed for visualization purposes.

=== PCITDC vs MOLE ===
{| class="wikitable"
|-
! scope="col" | Feature
! scope="col"| PCITDC
! scope="col"| MOLE
|-
! scope="row"| Detect Surface Cavities
| style="background-color: green; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Detect Interior Cavities
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Visualize Exact Channel Shape
| style="background-color: red; color: white;" | NO
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Works on Very Large Molecules
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO (very slow)
|-
! scope="row"| Works on Multi-Model Files
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Parallelization
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Relative Performance
| style="background-color: green; color: white;" | VERY FAST
| style="background-color: red; color: white;" | SLOW
|-
! scope="row"| Intuitive Visualization
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| False Positives
| style="background-color: orange; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Quantify Major Structural Features
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|}

=== Resources ===
* [https://blog.kitware.com/what-is-interpolatescalarsbeforemapping-in-vtk/ ParaView vertex colors]

== Resources ==

=== Software ===

* '''[https://www.rbvi.ucsf.edu/chimerax/ UCSF ChimeraX]''': Next-generation molecular visualization program from the Resource for Biocomputing, Visualization, and Informatics (RBVI).
* '''[https://www.ks.uiuc.edu/Research/vmd/ VMD]''': Molecular visualization program for displaying, animating, and analyzing large biomolecular systems using 3D graphics and built-in scripting. Used in conjunction with NAMD for visualizing molecular dynamics simulations.
* '''[https://www.ks.uiuc.edu/Research/namd/ NAMD]''': Parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. Scales to hundreds of cores for typical simulations and beyond 500,000 cores for the largest simulations.
* '''[https://webchem.ncbr.muni.cz/Platform/App/Mole MOLE]''': Universal toolkit for rapid location and characterization of channels and pores in biomacromolecules.
* '''[https://caver.cz/ CAVER]''': Software tool for the analysis and visualization of tunnels and channels in protein structures.
* '''[https://www.rcsb.org/ RCSB PDB Protein Data Bank]''': The largest and most popular database of protein structures.

=== Papers ===

* '''[https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5839519/ Survey: Geometric Protein Cavity Detection Algorithms]''': Paper surveying the current geometric algorithms for protein cavity detection. This is a great resource for understanding this project.
* '''[https://jcheminf.biomedcentral.com/articles/10.1186/1758-2946-5-39 MOLE 2.0]''': Paper describing the MOLE 2.0 protein cavity detection software. Contains a good algorithm that can be used for extracting a good approximation of the surface of a point cloud of atoms.
* '''[https://reader.elsevier.com/reader/sd/pii/S0969212607003759?token=4D0F6192E971D4406E8F7CBA047B0490C8715EBD67579D26427E0E63BD0298BEBF948DB238A28E4EAE2D17EBC6DF4E50 MOLE 1.0]''': Paper describing the MOLE 1.0 algorithm.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/epdf/10.1002/prot.22052 MolAxis]''': Paper describing the MolAxis protein cavity detection software.
* '''[https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002708 CAVER 3.0]''': Paper describing the CAVER 3.0 protein cavity detection software. This algorithm specializes in identifying tunnels.
* '''[https://www.sciencedirect.com/science/article/abs/pii/S0022283606008746 Travel Depth]''': Paper describing the "travel depth" algorithm for protein cavity visualization.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/full/10.1111/cgf.12120 GPU Protein Cavity Detection using Ambient Occlusion]''': Paper describing a GPU-based protein cavity detection algorithm that utilizes Ambient Occlusion.
* '''[https://www.aaai.org/Papers/KDD/1996/KDD96-037.pdf DBSCAN]''': Paper describing the DBSCAN clustering algorithm.
* '''[https://link.springer.com/article/10.1023/A:1009884809343 PDBSCAN]''': Paper describing a parallel version of DBSCAN.

=== Organizations/Groups ===

* '''[https://www.ks.uiuc.edu/ University of Illinois at Urbana-Champaign: Theoretical and Computational Biophysics Group]''': Perhaps the most important group in molecular dynamics. Developers of VMD and NAMD. Ran the simulation for the 200 million atom model of COVID-19.
* '''[https://www.ornl.gov/ Oak Ridge National Laboratory (ORNL)]''': Major national lab with active involvement in drug discovery and design using supercomputers.
* '''[http://rbvi.ucsf.edu/ Resource for Biocomputing, Visualization, and Informatics (RBVI)]''': Developers of UCSF Chimera and UCSF ChimeraX.
* '''[https://ist.ac.at/en/research/edelsbrunner-group/ IST Austria: Edelsbrunner Group]''': Important group in algorithms, computational geometry, and computational topology. Involved with protein cavity detection in the past and potentially involved now. The head of the group, Edelsbrunner created alpha shapes.
* '''[https://www.tacc.utexas.edu/systems/frontera Texas Advanced Computing Center]''': Maintainers of the Frontera supercomputer, the fifth fastest supercomputer in the world (2020). The simulation for the 200 million atom model of COVID-19 was executed on this supercomputer.
* '''[https://webchem.ncbr.muni.cz/ WebChemistry]''': Developed MOLE and many other software tools related to molecular analysis.

=== Important Links ===

* '''[https://www.hpcwire.com/off-the-wire/coronavirus-massive-simulations-completed-on-frontera-supercomputer/ COVID-19 200 million atom NAMD simulation]'''
* '''[https://www.ornl.gov/news/early-research-existing-drug-compounds-supercomputing-could-combat-coronavirus/ ORNL COVID-19 spike drug identification]'''

=== Proteins ===

* '''[https://www.rcsb.org/structure/1jj2 Large Ribosomal Subunit (98,693 atoms)]'''
* '''[https://www.rcsb.org/structure/6VXX COVID-19 Spike Glycoprotein, Closed State (23,694 atoms)]'''
* '''[https://www.rcsb.org/structure/6Y2E COVID-19 Main Protease (5,370 atoms)]'''
* '''[https://www.rcsb.org/structure/6VWW COVID-19 NSP15 Endoribonuclease (24,036 atoms)]'''

== References ==
[1] Simões, T., Lopes, D., Dias, S., Fernandes, F., Pereira, J., Jorge, J., Bajaj, C., & Gomes, A. (2017). Geometric Detection Algorithms for Cavities on Protein Surfaces in Molecular Graphics: A Survey. Computer graphics forum : journal of the European Association for Computer Graphics, 36(8), 643–683. https://doi.org/10.1111/cgf.13158

[2] Yaffe, Fishelovitch, Wolfson, Halperin, & Nussinov. (2008). MolAxis: Efficient and accurate identification of channels in macromolecules. Proteins: Structure, Function, and Bioinformatics, 73(1), 72-86.

[3] Coleman, R. G., & Sharp, K. A. (2006). Travel depth, a new shape descriptor for macromolecules: application to ligand binding. Journal of molecular biology, 362(3), 441–458. https://doi.org/10.1016/j.jmb.2006.07.022

Internal:GPCD

2020-07-21T16:15:59Z

Angelos: /* Protein Energetics */

== Introduction ==

=== Protein Cavities ===
Cavities are structural features of proteins and they are often important for drug discovery and design. There are different kinds of cavities, including pockets, voids, tunnels, and channels. The types of cavities are listed in more detail in figure 1.

[[File:proteinCavityTypes.jpg|center|frame| Figure 1: The basic types of protein cavities [1]]]

# '''Void''': A cavity that looks like internal "hole" in a protein with no connection to the surface.
# '''Cleft''': A pocket that looks like a "dent" in the surface of a protein. Can also be found as part of a tunnel or channel.
# '''Invagination''': A pocket that looks like a "cove". It has a small opening on the surface of a protein and leads to a larger interior. Can also be found as part of a tunnel or after a cleft.
# '''Tunnel''': A pocket that has an opening on the surface of a protein and leads to somewhere in the interior. Has no opening on the other side.
# '''Channel''': A cavity that has an opening on one side of the surface of a protein and leads to another opening in another location on the surface of the protein.

Of course, the types of cavities listed in figure 1 are only the basic ones. Combinations of these are also possible as shown in figure 2.

[[File:proteinCavityAdvancedTypes.jpg|center|frame| Figure 2: Combination types of protein cavities [1]]]

=== Protein Cavity Detection Methods ===
There are several methods for detecting protein cavities. There are sphere based methods, grid based methods, grid-and-sphere based methods, surface based methods, grid-and-surface based methods, time-varying methods, GPU based methods, consensus based methods, and finally tessellation based methods. These algorithms are shown in more detail in figure 3. Tessellation based methods mainly rely on Voronoi tessellation. Some methods rely on weighted Voronoi tessellation because of its ability to more accurately represent the volume of atoms. The Voronoi tessellation in these methods is computed from a set of points representing the atomic centers of the underlying protein.

[[File:proteinCavityDetectionAlgorithms.jpg|center|frame| Figure 3: A hierarchy of protein cavity detection algorithms [1]]]

== Literature Review ==

=== MOLE ===

==== Method ====
The MOLE algorithm is specifically designed to identify tunnels and channels. Finding superficial cavities is a secondary concern and does so as part of the surface approximation step. MOLE utilizes the surface obtained by the refinement of the convex hull to find superficial cavities such as clefts and invaginations. In essence, the approximation of a more accurate surface and the detection of these superficial cavities are closely related. As tetrahedra are removed from the convex hull to get a more accurate surface approximation, a parameter called ''interior threshold'' controls how many superficial cavities are found. For example, if the number of tetrahedra removed from a specific part of the molecule exceeds this threshold, the algorithm marks these tetrahedra as part of a superficial cavity.

[[File:moleVoronoiTunnelPath.jpg|center|frame| Figure 4: Shows in 2D the Voronoi diagram (thin lines) and the convex hull (dotted line) of molecule represented by atoms (circles). Each Voronoi edge is evaluated by the cost function (numbers). The thick line represents the optimal path found by the Dijkstra's algorithm from the given starting point (black circle), i.e., centerline of found tunnel. [1]]]

For the more complicated types of cavities, tunnels and channels, MOLE uses the Voronoi diagram to identify them. Start and end points are defined as the centers of the deepest tetrahedra in each cavity. The depth of the tetrahedron is defined as the number of Voronoi edges from the closest boundary tetrahedron (on the surface). Tunnels/channels are computed as the shortest paths between all pairs of start and end points in the same cavity diagram.

[[File:MoleDijkstraWeightFunction.png|center|frame| Figure 5: Shows the weight function used to estimate the weight of each Voronoi edge for Dijkstra's algorihtm. l(e) is the length of the edge, d(e) is the distance of the edge to the closest atom Van der Waals sphere and ϵ is a small number to avoid division by zero [1]]]

==== Complexity ====
* Worst-case: O(N^2 log N), where N is the number of atoms in the molecule.
* Average: O(M log M), where M is the number of vertices in the Voronoi diagram (in the worst case, M = N^2, but in most cases M = O(N)). As a result of the use of the incremental algorithm and Hilbert curve ordering, the complexity of calculating the Delaunay triangulation of most molecular structures is O(N log N). Finally, the complexity of all the remaining steps of the algorithm is at most O(M log M).

=== MolAxis ===

==== Method ====
The MolAxis algorithm relies on Voronoi and Dijkstra's algorithm to find channels, similarly to MOLE. The algorithm consists of the following high-level steps:

# Approximate the molecule using a collection of '''fixed-size spheres'''. The number of balls needed depends on the required quality of the approximation. In the extreme case, where the user is interested in a high resolution, each atom is approximated using a single sphere.
# Compute the '''Voronoi diagram''' of the centers of the fixed-sized spheres. The diagram is intersected with a user-defined bounding sphere. This defines the area of interest.
# Discard the parts of the Voronoi diagram that intersect the molecule.
# Define a '''graph''', using the Voronoi vertices and edges. A weight is given to each edge, which depends on its length and on its distance from the surface of the molecule.
# Define a '''source point (root vertex)'''. The user can either supply a source point or allow MolAxis to automatically choose the source point to be the '''center of the largest cavity in the molecule''' (basically detect channels inside-out).
# Apply a shortest path optimization algorithm ('''Dijktra's algorithm''') on the graph, computing a tree called the "corridor tree".
# The desired channels are paths in the corridor tree. Duplicate channels are avoided by using the split distance parameter. The split distance bounds the distance to the least common ancestor of the channel and all previously reported channels.

[[File:MolAxisVoronoi.png|center|frame| Figure 6: Shows an example of fixed-size circles (light-blue) approximating varying-size circles and the pathway diagram of their centers. The discarded portion of the Voronoi diagram of their centers is depicted using dotted lines [2]. This figure also helps illustrate the usefulness of the Voronoi tessellation in computing pathways.]]

== Protein Energetics ==
=== Metrics ===
# '''Hydrophobicity''': Physical property of a molecule that is seemingly repelled from a mass of water. Hydrophilic molecules are attracted to water. Hydrophobicity plays a major role in protein folding. Hydrophobic molecules also tend to be nonpolar and prefer other neutral molecules and nonpolar solvents. Hydrophobicity is the most widely used metric.
# '''Hydrogen Bonds''': Weak bonds between two molecules resulting from an electrostatic attraction between a proton in one molecule and an electronegative atom in the other. We often see these bonds between oxygen and hydrogen.
# '''Van der Waals Interactions''': Also known as Van der Waals forces, they are forces driven by induced electrical interactions between two or more atoms or molecules that are very close to each other. Van der Waals interaction is the weakest of all intermolecular attractions between molecules. These forces are distance-dependent.
# '''Salt Bridges''': Non-covalent (no sharing of eletrons; electromagnetic) interaction between two ionized sites. It has two components: a hydrogen bond and an electrostatic interaction.
# '''Disulfide Bridges''': Strong bonds that can form between two cysteines (an amino acid). The strength of disulfide bridges helps stabilize a protein. Disulfide bridges are especially common in proteins that are secreted from cells.
# '''Charge''': Plays a role in strength of bonds. A higher charge means that an atom will have stronger forces in an electromagnetic field.
# '''Polarity''': Separation of electric charge leading to a molecule or its chemical groups having an electric dipole moment, with a negatively charged end and a positively charged end. Indicates the existence of polar bonds, which are covalent (sharing of electrons) and are quite strong. About [https://chem.libretexts.org/Bookshelves/Inorganic_Chemistry/Book%3A_Introduction_to_Inorganic_Chemistry/01%3A_Review_of_Chemical_Bonding/1.04%3A_Bond_Polarity_and_Bond_Strength bond polarity].

== Possible Methods that need to be Explored ==

=== 1. PODM Tessellation Using Van der Waals Surface Image (Angelos) ===

The first method that could be used for detecting protein cavities in parallel involves using PODM to create a tessellation of an image representation of the Van der Waals surface of a protein. The Van der Waals surface is an accurate representation of the surface of a protein and represents each atom as a sphere of a radius corresponding to its Van der Waals radius.

The steps for this method are outlined below:

# '''Compute the Van der Waals surface (Angelos)''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface (Angelos)''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again necessary.
# '''Compute Delaunay tessellation using PODM''': Once we have the image of the Van der Waals surface, we use PODM to compute the Delaunay tessellation in parallel. A low enough delta should be used to ensure that the surface is as accurately captured as possible.
# '''Compute Voronoi tessellation using PODM''': The next step is to compute the Voronoi tessellation as the dual of the Delaunay tessellation in parallel. The Voronoi tessellation is useful for quickly identifying tunnels using Dijkstra's algorithm.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=400px>
File:6vxxVanDerWaalsSurface.png | thumb | Figure 7: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).

File:COVID-19-Spike-Glycoprotein-6vxx,d=5.png | thumb | Figure 8: A tessellation based on the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). A highly-accurate surface tessellation is produced.
</gallery>

=== 2. PDT Tessellation Using Point Cloud of Atoms (Spiros) ===

The second method involves giving PDT/PODM a point cloud of atoms and creating a tessellation from that. This would require that the Delaunay triangulation is generated and a method for computing an accurate surface for the point cloud is employed. Computing this surface is one of the most challenging steps of this method. A method that could be employed is α-shapes, but this has the disadvantage of not guaranteeing connectivity between tetrahedra. Another method probably needs to be used that can guarantee connectivity between the tetrahedra (such as the one from MOLE 2.5).

The steps for this method are outlined below:

# '''Read PDB file and make point cloud of atoms (Angelos)''': The first step would be to read an input PDB file for a protein and create a point cloud of atoms. This involves either writing a basic PDB file reader or finding one we can use. However, writing a basic one is better because it skips the overhead of other full-featured readers as we only care about the atomic centers and potentially the type of atoms.
# '''Compute Delaunay triangulation using PDT/PODM''': The second step is to compute the Delaunay triangulation from the point cloud in parallel. This will produce an inaccurate surface that needs to be refined.
# '''Compute accurate surface (Spiros)''': The Delaunay triangulation on a point cloud of atoms produces a surface that does not reflect the actual surface of the protein. As a result, an accurate surface needs to computed by refining the triangulation. This would involve removing tetrahedra until an accurate surface is found. This should be done in parallel.
## '''MOLE method''':
### The molecular surface is approximated by iterative removal of boundary tetrahedra from the outermost layers (i.e., tetrahedra found at the interface between the molecule and the external environment).
### Boundary tetrahedra produced by the triangulation are removed in this step if they are sufficiently large to fit a sphere with a given probe radius
#### Note: A tetrahedron T fits a sphere S with probe radius r if the center C of sphere S can be placed inside the tetrahedron and the distance to all vertices of T is greater than or equal to the sum of r, with the Van der Waals radius of an atom corresponding to the given vertex.
##### Explanation: Given a boundary tetrahedron, if a Sphere S with center = Incenter C, and radius = probe radius
### Next, tetrahedra that are too small to fit a sphere with interior radius are removed.
#### Explanation: Given a non-boundary tetrahedron, if the radius of its inscribed sphere is smaller than (probe?) radius, then the tetrahedron needs to be removed.
### Remaining tetrahedra form one or more connected components. We call the components that contain at least one tetrahedron on the molecular surface cavity diagrams.
### It should be noted here that the cavity diagram is a purely geometrical concept to help identify regions of space (volume) that can contain tunnels and only very loosely corresponds to the cavities shown in Figure 1B).
# '''Compute Voronoi tesselation''': Once the accurate surface is determined the Voronoi tessellation must be computed. This step could be done before the accurate surface is computed, but we would also have to refine the Voronoi if that is to happen.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=300px>

File:6vxxPointCloud.jpeg|thumb| Figure 9: The atomic point cloud of the COVID-19 spike glycoprotein (6vxx). There are 23,694 atoms (points) in this image.

File:6vxxUnrefinedDelaunay.jpeg|thumb| Figure 10: An unrefined Delaunay triangulation of the COVID-19 spike glycoprotein (6vxx). Generated using the Delaunay3D filter in ParaView.

File:6vxxAlphaShapesDelaunay.jpeg|thumb| Figure 11: A Delaunay triangulation using α-shapes with α=3 of the COVID-19 spike glycoprotein (6vxx). Using α-shapes we can get a better approximation of the surface. Generated using the Delaunay3D filter in ParaView.
</gallery>

=== 3. Surface-Grid-based Using Van der Waals Surface Volumetric Image ===

The third method involves writing a new software for doing protein cavity detection. This software would use a volumetric image (e.g. of type .mha) representing the Van der Waals surface of a protein. Uses perhaps the most faithful representation of the Van der Waals surface, which is the ground truth for a molecular surface. As long as the Van der Waal surface generation algorithm is fast, this method has the potential of being both accurate and fast as it eliminates the overhead of tessellation. Good parallelization is important for this method because there are many more voxels than there are tetrahedra in a tessellation-based method. Because this method operates on voxels, it is a candidate for future GPU parallelization. This method is also a strong candidate for deep learning, much more than the tessellation based methods, because its output can easily be used for training a deep learning model in the future.

The steps for this method are outline below:

# '''Compute the Van der Waals surface''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again likely necessary in the future.
# '''Identify cavities in parallel''': Using the image voxels, the cavities now need to be identified in parallel (potentially using GPU parallelization in the future). It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as volumetric image''': The last step is to output the detected cavities as a volumetric important. It is important to only include the identified cavities in this image. This way, researchers can overlay the normal protein structure with the volume of the cavities and easily see where they are.

<gallery mode="packed" heights=250px>
File:vanDerWaalsSurfaceImage2.jpeg|center|thumb|800px| Figure 12: Volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). There are 15.5 million voxels in this volume.

File:ProteinCavityDetectionSurfaceMethodGridExample.png|center|thumb|800px| Figure 13: Shows an example grid for a slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). The grey grid represents the scalar field whereas the green grid represents possible partitioning of the grid for multithreading. Finer or coarser partitioning may be used depending on the degree of parallelism.

File:VanDerWaalsSurfaceVolumeAnim.gif|center|frame| Figure 14: An animation of a thicker slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). Potential cavities and voids are clearly visible.
</gallery>

== Challenges ==

There are several challenges that need to be addressed if we are to have a good parallel solution for detecting protein cavities.

=== 1. Expensive Pre-Processing ===

The parallel methods involving the Van der Waals surface, especially the one with the image input have expensive pre-processing. For large molecules, getting this surface can be slow because existing implementations are sequential and written in Python. We need to come up with a way to quickly generate this surface and its image representation. This pre-processing does not have to be done if we operated on the point cloud using method 2 described previously, leading to a major speedup in overall performance.

[[File:6vxxVanDerWaalsSurface.png|center|thumb|800px| Figure 15: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).]]

=== 2. Massive Parallelism (Stretch Goal) ===

The solution we will implement may need to support massive parallelism, ideally using MPI. The reason for this is that molecular dynamics simulations can potentially involve a massive number of atoms. If we consider the 200 million coronavirus model, then a simulation containing 5000 frames, would involve a total of 1 trillion atoms. If we are to detect the cavities in every frame in real-time, a massively parallel solution is necessary. There is no doubt about why 250,000 CPU cores were required for the COVID-19 simulation.

<gallery mode="packed" heights=400px>
File:molecularDynamicsSimulation1.gif|frame| Figure 16: An animation produced by a molecular dynamics simulation.

File:molecularDynamicsSimulation2.gif|frame| Figure 17: An animation produced by the COVID-19 molecular dynamics simulation.
</gallery>

== Candidate Method: Protein Cavity Identification using Triangulation, unsigned Distance, and Clustering (PCITDC) ==
The PCITDC method employs Delaunay triangulation to generate a coarse and fine surface of a protein (the fine surface is enclosed within the coarse surface). Then, the unsigned distance of the fine surface from the coarse surface is computed. This is done by iterating through every vertex (atom) in the fine surface and computing its distance from the closest point on the coarse surface. Finally, three-level clustering is employed to cluster atoms based on their distance values.

The algorithm contains the following high-level steps:
=== (Spiros) Compute Inaccurate and Accurate surfaces ===
# Inaccurate and Accurate Surface Candidates
## '''Delaunay/Voronoi Based Surfaces''':
### (Inaccurate) Convex Hull ('''CHS''') [TODO]
#### Advantages:
##### It's easy to calculate
#### Disadvantages:
##### The depth values need to be scaled properly in the Unsigned Distance step.
#### Note: Shared memory parallelism can be employed [TODO]
### (Inaccurate and Accurate) Alpha-Shape Surface ('''ASS'''): [DONE]
#### Computed using:
##### Delaunay Triangulation
##### Removing all tetrahedra of a DT that their circumradius is less than a.
#### Advantages:
##### Has a direct relation to the molecule
##### Has few points -> fast computations further in the pipeline
#### Disadvantages:
##### The surface does not help with identifying channels
##### '''alpha''' needs fine-tuning [TODO]
##### Many disconnected sub-complexes may be created
#### Note: parallelism can be employed [TODO]
### (Inaccurate and Accurate) Mole Surface ('''MS'''): [DONE]
#### Computed using:
##### Delaunay Triangulation
##### First, we remove the boundary tetrahedra. Then we remove all the neighbors of the removed tetrahedra, that '''none''' of their vertices satisfy the formula, '''circumradius < probe-radius + atomRadius'''. We do that in a BFS fashion till there is no tetrahedron to be removed. This method is similar to AS.
#### Can be improved by:
##### Removing slivers using scaled volume, as it was done in PODM3D [TODO]
##### Not removing elements that included at least 3 atomic bond edges [DONE]
##### Not removing elements that will result in an element have no neighbors [ELIMINATED]
##### Not removing elements that result in losing vertex of the boundary [ELIMINATED]
#### Advantages:
##### Has a direct relation to the molecule
##### Has few points -> fast computations further in the pipeline
#### Disadvantages:
##### The surface does not help with identifying channels
###### but we can learn from MOLE [TODO]
##### '''Probe radius''' needs fine-tuning [TODO]
#### Note: parallelism needs to be explored [TODO]
### Delaunay Triangulation is performed using Dynamic Parallel Delaunay Triangulation (PDT) [DONE]
#### Can be improved by:
##### Implementing a weighted Delaunay Triangulation based on atomic radius [TODO]
#### Note: parallelism can be employed [TODO]
## '''Space Filling (Van der Waals) Surfaces'''
### (Accurate) Solvent Excluded Surface ('''SES''') [Done]
#### Computed using:
##### [https://www.cgl.ucsf.edu/chimerax/ Chimera], Programmed in Python (Open-Source) [DONE]
##### [https://zhanglab.ccmb.med.umich.edu/EDTSurf/ EDT Surf], Programmed in C++ (Open-Source) [TODO]
##### [https://github.com/kylelutz/chemkit Chemkit]. Programmed in C++ (Open-Source) [TODO]
#### Advantages:
##### We can extract all cavities using it
#### Disadvantages:
##### Does not have a direct relation to the molecule
##### Has many points -> slow computations further in the pipeline
#### Can be improved by:
##### Clustering decimation which essentially reduces the number of vertices/faces which results in having a reasonable small surface which is a good representation of the molecular surface. [TODO]
##### Converting it to image and creating a course surface using PODM3D [TODO]
#### Note: parallelism needs to be explored [TODO]
##### [https://link.springer.com/article/10.1007/s00371-017-1397-2 Interactive GPU-based generation of solvent-excluded surfaces]
# Possible Combinations:
## Inaccurate = MS and Accurate = SES [Done]
### Advantages:
#### Accurate depth evaluation:
### Disadvantages:
#### slow computations further in the pipeline:
## Inaccurate = MS and Accurate = MS [Done]
### Advantages:
#### fast computations further in the pipeline:
### Disadvantages:
#### Channels may not be captured
#### Not as accurate as option (1)
## Inaccurate = CH and Accurate = SES [TODO]
## Inaccurate = CH and Accurate = MS [TODO]

=== (Angelos/Spiros) Compute unsigned distance of the accurate from the inaccurate surface ===
# Algorithm:
## We iterate through every vertex (atom) in the accurate surface and compute its distance from the closest point on the inaccurate surface. This distance indicates the ''depth'' of a specific vertex. We apply Logarithmic scaling.
# Computed using:
## (Angelos) Using vtkStaticCellLocator (octree) [DONE]
### Note: Shared memory parallelism can be used [TODO]
## (Angelos/Spiros) Using vtkSignedDistance (signed distance) [TODO]
## (Spiros) Using path tracing in Delaunay triangulation [TODO]
### Algorithmic idea
#### Find the central cell of the triangulation and then traverse cells using visibility walk and jump and walk techniques in order to minimize the distance with the given point.
#### Once you reach the boundary, you will have a set of possible cells which include the minimum point, and afterward, you compute it.
### Note: Shared memory parallelism can be possibly employed [TODO]

=== (Angelos/Spiros) Compute atom clusters based on unsigned distance values: ===
# (Angelos) A modified version of DBSCAN that finds neighbors based on not only their density but also their unsigned distance values.
## Variations:
### (Anglelos) One-level clustering [DONE]
### (Anglelos) Multi-level clustering to capture all the types of cavities
#### E.g. Three-level clustering. The first level consists of large surface cavities. The second level consists of small surface cavities. The third level consists of deep cavities (i.e. channels). [DONE]
## Can be improved by:
### Implementation-wise:
#### (Spiros) Using mean-cluster-distance value instead of first-cluster-point-distance value. [DONE]
### Parameters Tuning:
#### (Spiros/Angelos) Utilizing min/mean/max atom-radius & min/mean/max bondLength to define '''eps / search-radius''' [TODO]
#### (Spiros) Utilizing sistibution statistics of distances to define '''minimum points per cluster'''. [TODO]
#### (Angelos) Identify distance areas, e.g. (0-40), where cavities do not exist. [TODO]
## Note: parallelism can be employed, see [https://link.springer.com/article/10.1023/A:1009884809343 A Fast Parallel Clustering Algorithm for Large Spatial Databases PDBSCAN] [TODO]
# (Angelos/Spiros) [https://pdfs.semanticscholar.org/f730/194462fb6bfcd0b90d7f22a960ed58d77199.pdf A Novel Spatial Clustering Algorithm Based on Delaunay Triangulation] [TODO]
## Note: parallelism can be explored [TODO]
# (Angelos) [https://dl.acm.org/doi/abs/10.1145/304181.304187 OPTICS: ordering points to identify the clustering structure] [TODO]

=== (Angelos) Classify clusters based on the type of cavity they are: ===
# Algorithms:
## Unsigned distance based classification using mean distance cluster value [TODO]
### Note: Shared memory parallelism can be used [TODO]
## Analyze the shape of clusters to identify what type of cavity they are. [TODO]
### This can potentially be done using machine learning.
### Note: Shared memory parallelism can be used [TODO]

=== Output: ===
# Labeled fine surface [DONE]
# Labeled protein [TODO]
## Note: a Paraview plugin may be developed for visualization purposes.

=== PCITDC vs MOLE ===
{| class="wikitable"
|-
! scope="col" | Feature
! scope="col"| PCITDC
! scope="col"| MOLE
|-
! scope="row"| Detect Surface Cavities
| style="background-color: green; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Detect Interior Cavities
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Visualize Exact Channel Shape
| style="background-color: red; color: white;" | NO
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Works on Very Large Molecules
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO (very slow)
|-
! scope="row"| Works on Multi-Model Files
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Parallelization
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Relative Performance
| style="background-color: green; color: white;" | VERY FAST
| style="background-color: red; color: white;" | SLOW
|-
! scope="row"| Intuitive Visualization
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| False Positives
| style="background-color: orange; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Quantify Major Structural Features
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|}

=== Resources ===
* [https://blog.kitware.com/what-is-interpolatescalarsbeforemapping-in-vtk/ ParaView vertex colors]

== Resources ==

=== Software ===

* '''[https://www.rbvi.ucsf.edu/chimerax/ UCSF ChimeraX]''': Next-generation molecular visualization program from the Resource for Biocomputing, Visualization, and Informatics (RBVI).
* '''[https://www.ks.uiuc.edu/Research/vmd/ VMD]''': Molecular visualization program for displaying, animating, and analyzing large biomolecular systems using 3D graphics and built-in scripting. Used in conjunction with NAMD for visualizing molecular dynamics simulations.
* '''[https://www.ks.uiuc.edu/Research/namd/ NAMD]''': Parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. Scales to hundreds of cores for typical simulations and beyond 500,000 cores for the largest simulations.
* '''[https://webchem.ncbr.muni.cz/Platform/App/Mole MOLE]''': Universal toolkit for rapid location and characterization of channels and pores in biomacromolecules.
* '''[https://caver.cz/ CAVER]''': Software tool for the analysis and visualization of tunnels and channels in protein structures.
* '''[https://www.rcsb.org/ RCSB PDB Protein Data Bank]''': The largest and most popular database of protein structures.

=== Papers ===

* '''[https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5839519/ Survey: Geometric Protein Cavity Detection Algorithms]''': Paper surveying the current geometric algorithms for protein cavity detection. This is a great resource for understanding this project.
* '''[https://jcheminf.biomedcentral.com/articles/10.1186/1758-2946-5-39 MOLE 2.0]''': Paper describing the MOLE 2.0 protein cavity detection software. Contains a good algorithm that can be used for extracting a good approximation of the surface of a point cloud of atoms.
* '''[https://reader.elsevier.com/reader/sd/pii/S0969212607003759?token=4D0F6192E971D4406E8F7CBA047B0490C8715EBD67579D26427E0E63BD0298BEBF948DB238A28E4EAE2D17EBC6DF4E50 MOLE 1.0]''': Paper describing the MOLE 1.0 algorithm.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/epdf/10.1002/prot.22052 MolAxis]''': Paper describing the MolAxis protein cavity detection software.
* '''[https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002708 CAVER 3.0]''': Paper describing the CAVER 3.0 protein cavity detection software. This algorithm specializes in identifying tunnels.
* '''[https://www.sciencedirect.com/science/article/abs/pii/S0022283606008746 Travel Depth]''': Paper describing the "travel depth" algorithm for protein cavity visualization.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/full/10.1111/cgf.12120 GPU Protein Cavity Detection using Ambient Occlusion]''': Paper describing a GPU-based protein cavity detection algorithm that utilizes Ambient Occlusion.

=== Organizations/Groups ===

* '''[https://www.ks.uiuc.edu/ University of Illinois at Urbana-Champaign: Theoretical and Computational Biophysics Group]''': Perhaps the most important group in molecular dynamics. Developers of VMD and NAMD. Ran the simulation for the 200 million atom model of COVID-19.
* '''[https://www.ornl.gov/ Oak Ridge National Laboratory (ORNL)]''': Major national lab with active involvement in drug discovery and design using supercomputers.
* '''[http://rbvi.ucsf.edu/ Resource for Biocomputing, Visualization, and Informatics (RBVI)]''': Developers of UCSF Chimera and UCSF ChimeraX.
* '''[https://ist.ac.at/en/research/edelsbrunner-group/ IST Austria: Edelsbrunner Group]''': Important group in algorithms, computational geometry, and computational topology. Involved with protein cavity detection in the past and potentially involved now. The head of the group, Edelsbrunner created alpha shapes.
* '''[https://www.tacc.utexas.edu/systems/frontera Texas Advanced Computing Center]''': Maintainers of the Frontera supercomputer, the fifth fastest supercomputer in the world (2020). The simulation for the 200 million atom model of COVID-19 was executed on this supercomputer.
* '''[https://webchem.ncbr.muni.cz/ WebChemistry]''': Developed MOLE and many other software tools related to molecular analysis.

=== Important Links ===

* '''[https://www.hpcwire.com/off-the-wire/coronavirus-massive-simulations-completed-on-frontera-supercomputer/ COVID-19 200 million atom NAMD simulation]'''
* '''[https://www.ornl.gov/news/early-research-existing-drug-compounds-supercomputing-could-combat-coronavirus/ ORNL COVID-19 spike drug identification]'''

=== Proteins ===

* '''[https://www.rcsb.org/structure/1jj2 Large Ribosomal Subunit (98,693 atoms)]'''
* '''[https://www.rcsb.org/structure/6VXX COVID-19 Spike Glycoprotein, Closed State (23,694 atoms)]'''
* '''[https://www.rcsb.org/structure/6Y2E COVID-19 Main Protease (5,370 atoms)]'''
* '''[https://www.rcsb.org/structure/6VWW COVID-19 NSP15 Endoribonuclease (24,036 atoms)]'''

== References ==
[1] Simões, T., Lopes, D., Dias, S., Fernandes, F., Pereira, J., Jorge, J., Bajaj, C., & Gomes, A. (2017). Geometric Detection Algorithms for Cavities on Protein Surfaces in Molecular Graphics: A Survey. Computer graphics forum : journal of the European Association for Computer Graphics, 36(8), 643–683. https://doi.org/10.1111/cgf.13158

[2] Yaffe, Fishelovitch, Wolfson, Halperin, & Nussinov. (2008). MolAxis: Efficient and accurate identification of channels in macromolecules. Proteins: Structure, Function, and Bioinformatics, 73(1), 72-86.

[3] Coleman, R. G., & Sharp, K. A. (2006). Travel depth, a new shape descriptor for macromolecules: application to ligand binding. Journal of molecular biology, 362(3), 441–458. https://doi.org/10.1016/j.jmb.2006.07.022

Internal:GPCD

2020-07-21T16:09:30Z

Angelos:

== Introduction ==

=== Protein Cavities ===
Cavities are structural features of proteins and they are often important for drug discovery and design. There are different kinds of cavities, including pockets, voids, tunnels, and channels. The types of cavities are listed in more detail in figure 1.

[[File:proteinCavityTypes.jpg|center|frame| Figure 1: The basic types of protein cavities [1]]]

# '''Void''': A cavity that looks like internal "hole" in a protein with no connection to the surface.
# '''Cleft''': A pocket that looks like a "dent" in the surface of a protein. Can also be found as part of a tunnel or channel.
# '''Invagination''': A pocket that looks like a "cove". It has a small opening on the surface of a protein and leads to a larger interior. Can also be found as part of a tunnel or after a cleft.
# '''Tunnel''': A pocket that has an opening on the surface of a protein and leads to somewhere in the interior. Has no opening on the other side.
# '''Channel''': A cavity that has an opening on one side of the surface of a protein and leads to another opening in another location on the surface of the protein.

Of course, the types of cavities listed in figure 1 are only the basic ones. Combinations of these are also possible as shown in figure 2.

[[File:proteinCavityAdvancedTypes.jpg|center|frame| Figure 2: Combination types of protein cavities [1]]]

=== Protein Cavity Detection Methods ===
There are several methods for detecting protein cavities. There are sphere based methods, grid based methods, grid-and-sphere based methods, surface based methods, grid-and-surface based methods, time-varying methods, GPU based methods, consensus based methods, and finally tessellation based methods. These algorithms are shown in more detail in figure 3. Tessellation based methods mainly rely on Voronoi tessellation. Some methods rely on weighted Voronoi tessellation because of its ability to more accurately represent the volume of atoms. The Voronoi tessellation in these methods is computed from a set of points representing the atomic centers of the underlying protein.

[[File:proteinCavityDetectionAlgorithms.jpg|center|frame| Figure 3: A hierarchy of protein cavity detection algorithms [1]]]

== Literature Review ==

=== MOLE ===

==== Method ====
The MOLE algorithm is specifically designed to identify tunnels and channels. Finding superficial cavities is a secondary concern and does so as part of the surface approximation step. MOLE utilizes the surface obtained by the refinement of the convex hull to find superficial cavities such as clefts and invaginations. In essence, the approximation of a more accurate surface and the detection of these superficial cavities are closely related. As tetrahedra are removed from the convex hull to get a more accurate surface approximation, a parameter called ''interior threshold'' controls how many superficial cavities are found. For example, if the number of tetrahedra removed from a specific part of the molecule exceeds this threshold, the algorithm marks these tetrahedra as part of a superficial cavity.

[[File:moleVoronoiTunnelPath.jpg|center|frame| Figure 4: Shows in 2D the Voronoi diagram (thin lines) and the convex hull (dotted line) of molecule represented by atoms (circles). Each Voronoi edge is evaluated by the cost function (numbers). The thick line represents the optimal path found by the Dijkstra's algorithm from the given starting point (black circle), i.e., centerline of found tunnel. [1]]]

For the more complicated types of cavities, tunnels and channels, MOLE uses the Voronoi diagram to identify them. Start and end points are defined as the centers of the deepest tetrahedra in each cavity. The depth of the tetrahedron is defined as the number of Voronoi edges from the closest boundary tetrahedron (on the surface). Tunnels/channels are computed as the shortest paths between all pairs of start and end points in the same cavity diagram.

[[File:MoleDijkstraWeightFunction.png|center|frame| Figure 5: Shows the weight function used to estimate the weight of each Voronoi edge for Dijkstra's algorihtm. l(e) is the length of the edge, d(e) is the distance of the edge to the closest atom Van der Waals sphere and ϵ is a small number to avoid division by zero [1]]]

==== Complexity ====
* Worst-case: O(N^2 log N), where N is the number of atoms in the molecule.
* Average: O(M log M), where M is the number of vertices in the Voronoi diagram (in the worst case, M = N^2, but in most cases M = O(N)). As a result of the use of the incremental algorithm and Hilbert curve ordering, the complexity of calculating the Delaunay triangulation of most molecular structures is O(N log N). Finally, the complexity of all the remaining steps of the algorithm is at most O(M log M).

=== MolAxis ===

==== Method ====
The MolAxis algorithm relies on Voronoi and Dijkstra's algorithm to find channels, similarly to MOLE. The algorithm consists of the following high-level steps:

# Approximate the molecule using a collection of '''fixed-size spheres'''. The number of balls needed depends on the required quality of the approximation. In the extreme case, where the user is interested in a high resolution, each atom is approximated using a single sphere.
# Compute the '''Voronoi diagram''' of the centers of the fixed-sized spheres. The diagram is intersected with a user-defined bounding sphere. This defines the area of interest.
# Discard the parts of the Voronoi diagram that intersect the molecule.
# Define a '''graph''', using the Voronoi vertices and edges. A weight is given to each edge, which depends on its length and on its distance from the surface of the molecule.
# Define a '''source point (root vertex)'''. The user can either supply a source point or allow MolAxis to automatically choose the source point to be the '''center of the largest cavity in the molecule''' (basically detect channels inside-out).
# Apply a shortest path optimization algorithm ('''Dijktra's algorithm''') on the graph, computing a tree called the "corridor tree".
# The desired channels are paths in the corridor tree. Duplicate channels are avoided by using the split distance parameter. The split distance bounds the distance to the least common ancestor of the channel and all previously reported channels.

[[File:MolAxisVoronoi.png|center|frame| Figure 6: Shows an example of fixed-size circles (light-blue) approximating varying-size circles and the pathway diagram of their centers. The discarded portion of the Voronoi diagram of their centers is depicted using dotted lines [2]. This figure also helps illustrate the usefulness of the Voronoi tessellation in computing pathways.]]

== Protein Energetics ==
=== Metrics ===
# '''Hydrophobicity''': Physical property of a molecule that is seemingly repelled from a mass of water. Hydrophilic molecules are attracted to water. Hydrophobicity plays a major role in protein folding. Hydrophobic molecules also tend to be nonpolar and prefer other neutral molecules and nonpolar solvents. Hydrophobicity is the most widely used metric.
# '''Hydrogen Bonds''': Weak bonds between two molecules resulting from an electrostatic attraction between a proton in one molecule and an electronegative atom in the other. We often see these bonds between oxygen and hydrogen.
# '''Van der Waals Interactions''': Also known as Van der Waals forces, they are forces driven by induced electrical interactions between two or more atoms or molecules that are very close to each other. Van der Waals interaction is the weakest of all intermolecular attractions between molecules. These forces are distance-dependent.
# '''Salt Bridges''': Non-covalent (no sharing of eletrons; electromagnetic) interaction between two ionized sites. It has two components: a hydrogen bond and an electrostatic interaction.
# '''Disulfide Bridges''': Strong bonds that can form between two cysteines (an amino acid). The strength of disulfide bridges helps stabilize a protein. Disulfide bridges are especially common in proteins that are secreted from cells.
# '''Charge''': Plays a role in strength of bonds. A higher charge means that an atom will have stronger forces in an electromagnetic field.
# '''Polarity''': Separation of electric charge leading to a molecule or its chemical groups having an electric dipole moment, with a negatively charged end and a positively charged end.

== Possible Methods that need to be Explored ==

=== 1. PODM Tessellation Using Van der Waals Surface Image (Angelos) ===

The first method that could be used for detecting protein cavities in parallel involves using PODM to create a tessellation of an image representation of the Van der Waals surface of a protein. The Van der Waals surface is an accurate representation of the surface of a protein and represents each atom as a sphere of a radius corresponding to its Van der Waals radius.

The steps for this method are outlined below:

# '''Compute the Van der Waals surface (Angelos)''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface (Angelos)''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again necessary.
# '''Compute Delaunay tessellation using PODM''': Once we have the image of the Van der Waals surface, we use PODM to compute the Delaunay tessellation in parallel. A low enough delta should be used to ensure that the surface is as accurately captured as possible.
# '''Compute Voronoi tessellation using PODM''': The next step is to compute the Voronoi tessellation as the dual of the Delaunay tessellation in parallel. The Voronoi tessellation is useful for quickly identifying tunnels using Dijkstra's algorithm.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=400px>
File:6vxxVanDerWaalsSurface.png | thumb | Figure 7: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).

File:COVID-19-Spike-Glycoprotein-6vxx,d=5.png | thumb | Figure 8: A tessellation based on the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). A highly-accurate surface tessellation is produced.
</gallery>

=== 2. PDT Tessellation Using Point Cloud of Atoms (Spiros) ===

The second method involves giving PDT/PODM a point cloud of atoms and creating a tessellation from that. This would require that the Delaunay triangulation is generated and a method for computing an accurate surface for the point cloud is employed. Computing this surface is one of the most challenging steps of this method. A method that could be employed is α-shapes, but this has the disadvantage of not guaranteeing connectivity between tetrahedra. Another method probably needs to be used that can guarantee connectivity between the tetrahedra (such as the one from MOLE 2.5).

The steps for this method are outlined below:

# '''Read PDB file and make point cloud of atoms (Angelos)''': The first step would be to read an input PDB file for a protein and create a point cloud of atoms. This involves either writing a basic PDB file reader or finding one we can use. However, writing a basic one is better because it skips the overhead of other full-featured readers as we only care about the atomic centers and potentially the type of atoms.
# '''Compute Delaunay triangulation using PDT/PODM''': The second step is to compute the Delaunay triangulation from the point cloud in parallel. This will produce an inaccurate surface that needs to be refined.
# '''Compute accurate surface (Spiros)''': The Delaunay triangulation on a point cloud of atoms produces a surface that does not reflect the actual surface of the protein. As a result, an accurate surface needs to computed by refining the triangulation. This would involve removing tetrahedra until an accurate surface is found. This should be done in parallel.
## '''MOLE method''':
### The molecular surface is approximated by iterative removal of boundary tetrahedra from the outermost layers (i.e., tetrahedra found at the interface between the molecule and the external environment).
### Boundary tetrahedra produced by the triangulation are removed in this step if they are sufficiently large to fit a sphere with a given probe radius
#### Note: A tetrahedron T fits a sphere S with probe radius r if the center C of sphere S can be placed inside the tetrahedron and the distance to all vertices of T is greater than or equal to the sum of r, with the Van der Waals radius of an atom corresponding to the given vertex.
##### Explanation: Given a boundary tetrahedron, if a Sphere S with center = Incenter C, and radius = probe radius
### Next, tetrahedra that are too small to fit a sphere with interior radius are removed.
#### Explanation: Given a non-boundary tetrahedron, if the radius of its inscribed sphere is smaller than (probe?) radius, then the tetrahedron needs to be removed.
### Remaining tetrahedra form one or more connected components. We call the components that contain at least one tetrahedron on the molecular surface cavity diagrams.
### It should be noted here that the cavity diagram is a purely geometrical concept to help identify regions of space (volume) that can contain tunnels and only very loosely corresponds to the cavities shown in Figure 1B).
# '''Compute Voronoi tesselation''': Once the accurate surface is determined the Voronoi tessellation must be computed. This step could be done before the accurate surface is computed, but we would also have to refine the Voronoi if that is to happen.
# '''Identify cavities in parallel (Angelos & Spiros)''': With the Delaunay and Voronoi tessellations computed, the cavities now need to be identified in parallel. It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as tessellations''': The last step is to output the detected cavities as a tessellated mesh. It is important to only include the identified cavities in this mesh by representing them as tetrahedra. This way, researchers can overlay the normal protein structure with the tessellations of the cavities and easily see where they are.

<gallery mode="packed" heights=300px>

File:6vxxPointCloud.jpeg|thumb| Figure 9: The atomic point cloud of the COVID-19 spike glycoprotein (6vxx). There are 23,694 atoms (points) in this image.

File:6vxxUnrefinedDelaunay.jpeg|thumb| Figure 10: An unrefined Delaunay triangulation of the COVID-19 spike glycoprotein (6vxx). Generated using the Delaunay3D filter in ParaView.

File:6vxxAlphaShapesDelaunay.jpeg|thumb| Figure 11: A Delaunay triangulation using α-shapes with α=3 of the COVID-19 spike glycoprotein (6vxx). Using α-shapes we can get a better approximation of the surface. Generated using the Delaunay3D filter in ParaView.
</gallery>

=== 3. Surface-Grid-based Using Van der Waals Surface Volumetric Image ===

The third method involves writing a new software for doing protein cavity detection. This software would use a volumetric image (e.g. of type .mha) representing the Van der Waals surface of a protein. Uses perhaps the most faithful representation of the Van der Waals surface, which is the ground truth for a molecular surface. As long as the Van der Waal surface generation algorithm is fast, this method has the potential of being both accurate and fast as it eliminates the overhead of tessellation. Good parallelization is important for this method because there are many more voxels than there are tetrahedra in a tessellation-based method. Because this method operates on voxels, it is a candidate for future GPU parallelization. This method is also a strong candidate for deep learning, much more than the tessellation based methods, because its output can easily be used for training a deep learning model in the future.

The steps for this method are outline below:

# '''Compute the Van der Waals surface''': This polygonal surface can be computed using UCSF Chimera. The output from this is an STL file containing the polygonal surface mesh. However, because the implementation is sequential, this step can present a major bottleneck in the overall method. In order to speed the process up, a custom parallel implementation to generate this surface will be necessary in the future.
# '''Generate image of the Van der Waals surface''': Because PODM requires an image as input, the polygonal Van der Waals surface needs to be converted to an image. This is done by "filling" the polygonal surface with a constant value. The segmentation module of 3D Slicer can be used for doing this, but as before, this step can present a major bottleneck because the implementation is sequential. A custom parallel implementation is once again likely necessary in the future.
# '''Identify cavities in parallel''': Using the image voxels, the cavities now need to be identified in parallel (potentially using GPU parallelization in the future). It is crucial that this step is done in parallel as no other method implements "proper" parallelization. This is by far the most challenging step.
# '''Output detected cavities as volumetric image''': The last step is to output the detected cavities as a volumetric important. It is important to only include the identified cavities in this image. This way, researchers can overlay the normal protein structure with the volume of the cavities and easily see where they are.

<gallery mode="packed" heights=250px>
File:vanDerWaalsSurfaceImage2.jpeg|center|thumb|800px| Figure 12: Volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). There are 15.5 million voxels in this volume.

File:ProteinCavityDetectionSurfaceMethodGridExample.png|center|thumb|800px| Figure 13: Shows an example grid for a slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). The grey grid represents the scalar field whereas the green grid represents possible partitioning of the grid for multithreading. Finer or coarser partitioning may be used depending on the degree of parallelism.

File:VanDerWaalsSurfaceVolumeAnim.gif|center|frame| Figure 14: An animation of a thicker slice of the volumetric image of the Van der Waals surface of the COVID-19 spike glycoprotein (6vxx). Potential cavities and voids are clearly visible.
</gallery>

== Challenges ==

There are several challenges that need to be addressed if we are to have a good parallel solution for detecting protein cavities.

=== 1. Expensive Pre-Processing ===

The parallel methods involving the Van der Waals surface, especially the one with the image input have expensive pre-processing. For large molecules, getting this surface can be slow because existing implementations are sequential and written in Python. We need to come up with a way to quickly generate this surface and its image representation. This pre-processing does not have to be done if we operated on the point cloud using method 2 described previously, leading to a major speedup in overall performance.

[[File:6vxxVanDerWaalsSurface.png|center|thumb|800px| Figure 15: The Van der Waals surface of the COVID-19 spike glycoprotein (6vxx).]]

=== 2. Massive Parallelism (Stretch Goal) ===

The solution we will implement may need to support massive parallelism, ideally using MPI. The reason for this is that molecular dynamics simulations can potentially involve a massive number of atoms. If we consider the 200 million coronavirus model, then a simulation containing 5000 frames, would involve a total of 1 trillion atoms. If we are to detect the cavities in every frame in real-time, a massively parallel solution is necessary. There is no doubt about why 250,000 CPU cores were required for the COVID-19 simulation.

<gallery mode="packed" heights=400px>
File:molecularDynamicsSimulation1.gif|frame| Figure 16: An animation produced by a molecular dynamics simulation.

File:molecularDynamicsSimulation2.gif|frame| Figure 17: An animation produced by the COVID-19 molecular dynamics simulation.
</gallery>

== Candidate Method: Protein Cavity Identification using Triangulation, unsigned Distance, and Clustering (PCITDC) ==
The PCITDC method employs Delaunay triangulation to generate a coarse and fine surface of a protein (the fine surface is enclosed within the coarse surface). Then, the unsigned distance of the fine surface from the coarse surface is computed. This is done by iterating through every vertex (atom) in the fine surface and computing its distance from the closest point on the coarse surface. Finally, three-level clustering is employed to cluster atoms based on their distance values.

The algorithm contains the following high-level steps:
=== (Spiros) Compute Inaccurate and Accurate surfaces ===
# Inaccurate and Accurate Surface Candidates
## '''Delaunay/Voronoi Based Surfaces''':
### (Inaccurate) Convex Hull ('''CHS''') [TODO]
#### Advantages:
##### It's easy to calculate
#### Disadvantages:
##### The depth values need to be scaled properly in the Unsigned Distance step.
#### Note: Shared memory parallelism can be employed [TODO]
### (Inaccurate and Accurate) Alpha-Shape Surface ('''ASS'''): [DONE]
#### Computed using:
##### Delaunay Triangulation
##### Removing all tetrahedra of a DT that their circumradius is less than a.
#### Advantages:
##### Has a direct relation to the molecule
##### Has few points -> fast computations further in the pipeline
#### Disadvantages:
##### The surface does not help with identifying channels
##### '''alpha''' needs fine-tuning [TODO]
##### Many disconnected sub-complexes may be created
#### Note: parallelism can be employed [TODO]
### (Inaccurate and Accurate) Mole Surface ('''MS'''): [DONE]
#### Computed using:
##### Delaunay Triangulation
##### First, we remove the boundary tetrahedra. Then we remove all the neighbors of the removed tetrahedra, that '''none''' of their vertices satisfy the formula, '''circumradius < probe-radius + atomRadius'''. We do that in a BFS fashion till there is no tetrahedron to be removed. This method is similar to AS.
#### Can be improved by:
##### Removing slivers using scaled volume, as it was done in PODM3D [TODO]
##### Not removing elements that included at least 3 atomic bond edges [DONE]
##### Not removing elements that will result in an element have no neighbors [ELIMINATED]
##### Not removing elements that result in losing vertex of the boundary [ELIMINATED]
#### Advantages:
##### Has a direct relation to the molecule
##### Has few points -> fast computations further in the pipeline
#### Disadvantages:
##### The surface does not help with identifying channels
###### but we can learn from MOLE [TODO]
##### '''Probe radius''' needs fine-tuning [TODO]
#### Note: parallelism needs to be explored [TODO]
### Delaunay Triangulation is performed using Dynamic Parallel Delaunay Triangulation (PDT) [DONE]
#### Can be improved by:
##### Implementing a weighted Delaunay Triangulation based on atomic radius [TODO]
#### Note: parallelism can be employed [TODO]
## '''Space Filling (Van der Waals) Surfaces'''
### (Accurate) Solvent Excluded Surface ('''SES''') [Done]
#### Computed using:
##### [https://www.cgl.ucsf.edu/chimerax/ Chimera], Programmed in Python (Open-Source) [DONE]
##### [https://zhanglab.ccmb.med.umich.edu/EDTSurf/ EDT Surf], Programmed in C++ (Open-Source) [TODO]
##### [https://github.com/kylelutz/chemkit Chemkit]. Programmed in C++ (Open-Source) [TODO]
#### Advantages:
##### We can extract all cavities using it
#### Disadvantages:
##### Does not have a direct relation to the molecule
##### Has many points -> slow computations further in the pipeline
#### Can be improved by:
##### Clustering decimation which essentially reduces the number of vertices/faces which results in having a reasonable small surface which is a good representation of the molecular surface. [TODO]
##### Converting it to image and creating a course surface using PODM3D [TODO]
#### Note: parallelism needs to be explored [TODO]
##### [https://link.springer.com/article/10.1007/s00371-017-1397-2 Interactive GPU-based generation of solvent-excluded surfaces]
# Possible Combinations:
## Inaccurate = MS and Accurate = SES [Done]
### Advantages:
#### Accurate depth evaluation:
### Disadvantages:
#### slow computations further in the pipeline:
## Inaccurate = MS and Accurate = MS [Done]
### Advantages:
#### fast computations further in the pipeline:
### Disadvantages:
#### Channels may not be captured
#### Not as accurate as option (1)
## Inaccurate = CH and Accurate = SES [TODO]
## Inaccurate = CH and Accurate = MS [TODO]

=== (Angelos/Spiros) Compute unsigned distance of the accurate from the inaccurate surface ===
# Algorithm:
## We iterate through every vertex (atom) in the accurate surface and compute its distance from the closest point on the inaccurate surface. This distance indicates the ''depth'' of a specific vertex. We apply Logarithmic scaling.
# Computed using:
## (Angelos) Using vtkStaticCellLocator (octree) [DONE]
### Note: Shared memory parallelism can be used [TODO]
## (Angelos/Spiros) Using vtkSignedDistance (signed distance) [TODO]
## (Spiros) Using path tracing in Delaunay triangulation [TODO]
### Algorithmic idea
#### Find the central cell of the triangulation and then traverse cells using visibility walk and jump and walk techniques in order to minimize the distance with the given point.
#### Once you reach the boundary, you will have a set of possible cells which include the minimum point, and afterward, you compute it.
### Note: Shared memory parallelism can be possibly employed [TODO]

=== (Angelos/Spiros) Compute atom clusters based on unsigned distance values: ===
# (Angelos) A modified version of DBSCAN that finds neighbors based on not only their density but also their unsigned distance values.
## Variations:
### (Anglelos) One-level clustering [DONE]
### (Anglelos) Multi-level clustering to capture all the types of cavities
#### E.g. Three-level clustering. The first level consists of large surface cavities. The second level consists of small surface cavities. The third level consists of deep cavities (i.e. channels). [DONE]
## Can be improved by:
### Implementation-wise:
#### (Spiros) Using mean-cluster-distance value instead of first-cluster-point-distance value. [DONE]
### Parameters Tuning:
#### (Spiros/Angelos) Utilizing min/mean/max atom-radius & min/mean/max bondLength to define '''eps / search-radius''' [TODO]
#### (Spiros) Utilizing sistibution statistics of distances to define '''minimum points per cluster'''. [TODO]
#### (Angelos) Identify distance areas, e.g. (0-40), where cavities do not exist. [TODO]
## Note: parallelism can be employed, see [https://link.springer.com/article/10.1023/A:1009884809343 A Fast Parallel Clustering Algorithm for Large Spatial Databases PDBSCAN] [TODO]
# (Angelos/Spiros) [https://pdfs.semanticscholar.org/f730/194462fb6bfcd0b90d7f22a960ed58d77199.pdf A Novel Spatial Clustering Algorithm Based on Delaunay Triangulation] [TODO]
## Note: parallelism can be explored [TODO]
# (Angelos) [https://dl.acm.org/doi/abs/10.1145/304181.304187 OPTICS: ordering points to identify the clustering structure] [TODO]

=== (Angelos) Classify clusters based on the type of cavity they are: ===
# Algorithms:
## Unsigned distance based classification using mean distance cluster value [TODO]
### Note: Shared memory parallelism can be used [TODO]
## Analyze the shape of clusters to identify what type of cavity they are. [TODO]
### This can potentially be done using machine learning.
### Note: Shared memory parallelism can be used [TODO]

=== Output: ===
# Labeled fine surface [DONE]
# Labeled protein [TODO]
## Note: a Paraview plugin may be developed for visualization purposes.

=== PCITDC vs MOLE ===
{| class="wikitable"
|-
! scope="col" | Feature
! scope="col"| PCITDC
! scope="col"| MOLE
|-
! scope="row"| Detect Surface Cavities
| style="background-color: green; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Detect Interior Cavities
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Visualize Exact Channel Shape
| style="background-color: red; color: white;" | NO
| style="background-color: green; color: white;" | YES
|-
! scope="row"| Works on Very Large Molecules
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO (very slow)
|-
! scope="row"| Works on Multi-Model Files
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Parallelization
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|-
! scope="row"| Relative Performance
| style="background-color: green; color: white;" | VERY FAST
| style="background-color: red; color: white;" | SLOW
|-
! scope="row"| Intuitive Visualization
| style="background-color: green; color: white;" | YES
| style="background-color: green; color: white;" | YES
|-
! scope="row"| False Positives
| style="background-color: orange; color: white;" | YES (more)
| style="background-color: green; color: white;" | YES (less)
|-
! scope="row"| Quantify Major Structural Features
| style="background-color: green; color: white;" | YES
| style="background-color: red; color: white;" | NO
|}

=== Resources ===
* [https://blog.kitware.com/what-is-interpolatescalarsbeforemapping-in-vtk/ ParaView vertex colors]

== Resources ==

=== Software ===

* '''[https://www.rbvi.ucsf.edu/chimerax/ UCSF ChimeraX]''': Next-generation molecular visualization program from the Resource for Biocomputing, Visualization, and Informatics (RBVI).
* '''[https://www.ks.uiuc.edu/Research/vmd/ VMD]''': Molecular visualization program for displaying, animating, and analyzing large biomolecular systems using 3D graphics and built-in scripting. Used in conjunction with NAMD for visualizing molecular dynamics simulations.
* '''[https://www.ks.uiuc.edu/Research/namd/ NAMD]''': Parallel molecular dynamics code designed for high-performance simulation of large biomolecular systems. Scales to hundreds of cores for typical simulations and beyond 500,000 cores for the largest simulations.
* '''[https://webchem.ncbr.muni.cz/Platform/App/Mole MOLE]''': Universal toolkit for rapid location and characterization of channels and pores in biomacromolecules.
* '''[https://caver.cz/ CAVER]''': Software tool for the analysis and visualization of tunnels and channels in protein structures.
* '''[https://www.rcsb.org/ RCSB PDB Protein Data Bank]''': The largest and most popular database of protein structures.

=== Papers ===

* '''[https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5839519/ Survey: Geometric Protein Cavity Detection Algorithms]''': Paper surveying the current geometric algorithms for protein cavity detection. This is a great resource for understanding this project.
* '''[https://jcheminf.biomedcentral.com/articles/10.1186/1758-2946-5-39 MOLE 2.0]''': Paper describing the MOLE 2.0 protein cavity detection software. Contains a good algorithm that can be used for extracting a good approximation of the surface of a point cloud of atoms.
* '''[https://reader.elsevier.com/reader/sd/pii/S0969212607003759?token=4D0F6192E971D4406E8F7CBA047B0490C8715EBD67579D26427E0E63BD0298BEBF948DB238A28E4EAE2D17EBC6DF4E50 MOLE 1.0]''': Paper describing the MOLE 1.0 algorithm.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/epdf/10.1002/prot.22052 MolAxis]''': Paper describing the MolAxis protein cavity detection software.
* '''[https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1002708 CAVER 3.0]''': Paper describing the CAVER 3.0 protein cavity detection software. This algorithm specializes in identifying tunnels.
* '''[https://www.sciencedirect.com/science/article/abs/pii/S0022283606008746 Travel Depth]''': Paper describing the "travel depth" algorithm for protein cavity visualization.
* '''[https://onlinelibrary-wiley-com.proxy.lib.odu.edu/doi/full/10.1111/cgf.12120 GPU Protein Cavity Detection using Ambient Occlusion]''': Paper describing a GPU-based protein cavity detection algorithm that utilizes Ambient Occlusion.

=== Organizations/Groups ===

* '''[https://www.ks.uiuc.edu/ University of Illinois at Urbana-Champaign: Theoretical and Computational Biophysics Group]''': Perhaps the most important group in molecular dynamics. Developers of VMD and NAMD. Ran the simulation for the 200 million atom model of COVID-19.
* '''[https://www.ornl.gov/ Oak Ridge National Laboratory (ORNL)]''': Major national lab with active involvement in drug discovery and design using supercomputers.
* '''[http://rbvi.ucsf.edu/ Resource for Biocomputing, Visualization, and Informatics (RBVI)]''': Developers of UCSF Chimera and UCSF ChimeraX.
* '''[https://ist.ac.at/en/research/edelsbrunner-group/ IST Austria: Edelsbrunner Group]''': Important group in algorithms, computational geometry, and computational topology. Involved with protein cavity detection in the past and potentially involved now. The head of the group, Edelsbrunner created alpha shapes.
* '''[https://www.tacc.utexas.edu/systems/frontera Texas Advanced Computing Center]''': Maintainers of the Frontera supercomputer, the fifth fastest supercomputer in the world (2020). The simulation for the 200 million atom model of COVID-19 was executed on this supercomputer.
* '''[https://webchem.ncbr.muni.cz/ WebChemistry]''': Developed MOLE and many other software tools related to molecular analysis.

=== Important Links ===

* '''[https://www.hpcwire.com/off-the-wire/coronavirus-massive-simulations-completed-on-frontera-supercomputer/ COVID-19 200 million atom NAMD simulation]'''
* '''[https://www.ornl.gov/news/early-research-existing-drug-compounds-supercomputing-could-combat-coronavirus/ ORNL COVID-19 spike drug identification]'''

=== Proteins ===

* '''[https://www.rcsb.org/structure/1jj2 Large Ribosomal Subunit (98,693 atoms)]'''
* '''[https://www.rcsb.org/structure/6VXX COVID-19 Spike Glycoprotein, Closed State (23,694 atoms)]'''
* '''[https://www.rcsb.org/structure/6Y2E COVID-19 Main Protease (5,370 atoms)]'''
* '''[https://www.rcsb.org/structure/6VWW COVID-19 NSP15 Endoribonuclease (24,036 atoms)]'''

== References ==
[1] Simões, T., Lopes, D., Dias, S., Fernandes, F., Pereira, J., Jorge, J., Bajaj, C., & Gomes, A. (2017). Geometric Detection Algorithms for Cavities on Protein Surfaces in Molecular Graphics: A Survey. Computer graphics forum : journal of the European Association for Computer Graphics, 36(8), 643–683. https://doi.org/10.1111/cgf.13158

[2] Yaffe, Fishelovitch, Wolfson, Halperin, & Nussinov. (2008). MolAxis: Efficient and accurate identification of channels in macromolecules. Proteins: Structure, Function, and Bioinformatics, 73(1), 72-86.

[3] Coleman, R. G., & Sharp, K. A. (2006). Travel depth, a new shape descriptor for macromolecules: application to ligand binding. Journal of molecular biology, 362(3), 441–458. https://doi.org/10.1016/j.jmb.2006.07.022

File:Adaptivity PPoSS.png

2020-07-17T19:00:39Z

Angelos: Angelos uploaded a new version of File:Adaptivity PPoSS.png

File:Adaptivity PPoSS.png

2020-07-17T18:56:16Z

Angelos: Angelos uploaded a new version of File:Adaptivity PPoSS.png

File:Adaptivity PPoSS.png

2020-07-17T17:56:24Z

Angelos: Angelos uploaded a new version of File:Adaptivity PPoSS.png

Molecular Geometric Feature Analysis

2020-07-10T23:46:22Z

Angelos: /* SARS-CoV-2 Human Antibody Fab (Ab C105, PDB: 6XCM) */

== SARS-CoV-2 Antibodies ==
=== SARS-CoV-2 Human Antibody Fab (Ab C105, PDB: [https://www.rcsb.org/structure/6xcm 6XCM]) ===
<html>
<div style="text-align:center;">
<video width="1160" height="530" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/sars-cov-2-spike-antibody-contact.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 D614 ==
=== SARS-CoV-2 D614 Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 D614 Spike Cleavage Domain Tunnel Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614TunnelAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 G614 ==
=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Invagination Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614InvaginationAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 1 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel1Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 2 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel2Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 Glycoprotein Spike Geometric Features ==
<gallery heights=300px widths=300px mode="slideshow">
File:Protein-Slide19.jpg
File:Protein-Slide20.jpg ‎
File:Protein-Slide21.jpg
File:Protein-Slide22.jpg
File:Protein-Slide23.jpg
File:Protein-Slide24.jpg
</gallery>

Molecular Geometric Feature Analysis

2020-07-10T21:44:13Z

Angelos:

== SARS-CoV-2 Antibodies ==
=== SARS-CoV-2 Human Antibody Fab (Ab C105, PDB: 6XCM) ===
<html>
<div style="text-align:center;">
<video width="1160" height="530" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/sars-cov-2-spike-antibody-contact.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 D614 ==
=== SARS-CoV-2 D614 Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 D614 Spike Cleavage Domain Tunnel Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614TunnelAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 G614 ==
=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Invagination Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614InvaginationAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 1 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel1Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 2 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="700" height="500" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel2Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 Glycoprotein Spike Geometric Features ==
<gallery heights=300px widths=300px mode="slideshow">
File:Protein-Slide19.jpg
File:Protein-Slide20.jpg ‎
File:Protein-Slide21.jpg
File:Protein-Slide22.jpg
File:Protein-Slide23.jpg
File:Protein-Slide24.jpg
</gallery>

Main Page

2020-07-10T17:28:51Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strain

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-10T17:28:31Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|600px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) using the atomic structure from the original strain ====
<html>
<div style="text-align:center;">
<video width="700" height="500" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

'''Remark''': Notice the difference between the position of new tunnels where the C & O used to be in the original COVID19 strength

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Molecular Geometric Feature Analysis

2020-07-07T19:39:56Z

Angelos:

== SARS-CoV-2 D614 ==
=== SARS-CoV-2 D614 Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 D614 Spike Cleavage Domain Tunnel Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614TunnelAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 G614 ==
=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Pockets/Cavities Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614CavitiesAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Invagination Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614InvaginationAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 1 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel1Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

=== SARS-CoV-2 G614 Mutation Spike Cleavage Domain Tunnel 2 Close-Up Superimposed on Atomic Structure ===
<html>
<div style="text-align:center;">
<video width="868" height="580" controls loop>
<source src="https://crtc.cs.odu.edu/assets/videos/G614Tunnel2Anim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

== SARS-CoV-2 Glycoprotein Spike Geometric Features ==
<gallery heights=300px widths=300px mode="slideshow">
File:Protein-Slide19.jpg
File:Protein-Slide20.jpg ‎
File:Protein-Slide21.jpg
File:Protein-Slide22.jpg
File:Protein-Slide23.jpg
File:Protein-Slide24.jpg
</gallery>

Main Page

2020-07-07T19:34:15Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

To see other videos please click [[Molecular Geometric Feature Analysis | here]].

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-07T19:32:00Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 D614 vs G614 Mutation in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other videos and slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-07T19:31:41Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

==== SARS-CoV-2 G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

==== SARS-CoV-2 G614 vs D614 Mutations in Spike Cleavage Domain ====
<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

To see other videos and slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-07T19:30:36Z

Angelos: /* Announcements */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

==== G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

To see other videos and slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-07T19:29:35Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

==== G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls autoplay loop>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

To see previous slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

Main Page

2020-07-07T19:27:08Z

Angelos: /* Molecular Geometric Feature Analysis */

[[File:Thanksgiving lunch 2019.jpg|700px|thumb|frameless|center|CRTC Group: Giving thanks for a productive year with such a wonderful group, very proud to share this year with all of you at CRTC (Thanks Giving 2019).]]

== Announcements ==

===Molecular Geometric Feature Analysis===

<gallery heights=300px widths=300px mode="slideshow">
File:CRTC_ProteinCavities_D614G_1.png
File:CRTC_ProteinCavities_D614G_2.png ‎
File:CRTC_ProteinCavities_D614G_3.png
File:CRTC_ProteinCavities_D614G_4.png
File:CRTC_ProteinCavities_D614G_5.png
File:CRTC_ProteinCavities_D614G_6.png
File:CRTC_ProteinCavities_D614G_7.png
File:CRTC_ProteinCavities_D614G_8.png
</gallery>

==== G614 Spike Cleavage Domain Tunnels (red-blue) Superimposed on D614 Spike Cleavage Domain with D614 Tunnel (purple-orange) ====
<html>
<div style="text-align:center;">
<video width="868" height="580" controls>
<source src="https://crtc.cs.odu.edu/assets/videos/D614-G614-TunnelsSuperimposeAnim-web.mp4" type="video/mp4">
Your browser does not support the video tag.
</video>
</div>
</html>

To see previous slideshows please click [[Molecular Geometric Feature Analysis | here]].

=== Workshops ===
{| class="wikitable"
! scope="col" style="width: 250px;" | CNF HPC Workshop
! scope="col" style="width: 250px;" | CNF Imaging Workshop
|-
|The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a one-day mini-workshop on Imaging and High Performance Computing (HPC). For more information see: '''[[CNF_HPC_Workshop]]''' [[File:Logo-hpc.png|250px| center]]
|| The CRTC lab in collaboration with the [https://www.jlab.org/ Jefferson lab] hosted a two-day mini-workshop on Imaging and Nuclear Femptography. For more information see: '''[[CNF_Imaging_Workshop]]''' [[File:Cnf-workshop-logo-with-letters.png|250px| center]]

|}

== Introduction ==
Finite Element (FE) mesh generation is a critical component for many (bio-)engineering and science applications. CRTCLab is developing a novel framework for highly scalable and energy-efficient high-quality mesh generation for the Finite Element Analysis in three and four dimensions. CRTC effort and focus is on research activities that combine domain-and application-speciﬁc knowledge with run-time system support to improve energy efficiency and scalability of parallel FE mesh generation codes. Traditionally, parallel FE mesh generation methods and software are developed without considering the architectural features of the supercomputer platforms on which they are eventually used for production. The main reason is the complexity of sequential, and moreover parallel, mesh generation algorithms. As a result, it is too expensive, in terms of labor and time, to customize the performance of parallel mesh generation software for speciﬁc supercomputing architectures. CRTC’s basic research is funded from NSF, NASA, NRA, total $2M and target (bio-)engineering applications are: [[#Extreme-Scale_Parallel_Mesh_Generation_&_Adaptation | Aerospace Industry Applications]] and [[#Medical_Image_Computing | Health Care Industry Applications]].

<center>
'''<big>Principles for Building Practical and Scalable Domain-Specific Environment: Adaptive Data-Driven Mesh Generation for Modeling & Simulations</big>
'''
</center>
[[File:Adaptivity PPoSS.png|thumb|center|600px| This graph indicates the core enabling technologies we develop at CRTC: (i) Adaptive Tessellation Techniques: Theory and Systems and the corresponding and (ii) Parallel Runtime Software Systems required to run on current and emerging Supercomputers. The pictures around indicate the five main applications we target from left to right: (a) Computational Fluid Dynamics, (b) Numerical Flow Simulations used to Predict Flow Diversion Treatment Efficacy of Cerebral Aneurysms, (c) Trauma Brain Injuries, (d) Physic Based Adaptive Deformable Registration used in Image-Guided Neurosurgery, and (e) Imaging for Nuclear Femtography with our collaborators: from NASA/LARC, Stony Brook Medical School, Navy Research Lab, Harvard Medical School, and Physics and Jefferson Lab, respectively. The common denominator for all these applications is adaptive parallel tessellation ( or known in the science and engineering community as mesh generation or image-to-mesh conversion). Parallel mesh generation is "a new research area between the boundaries of two [https://en.wikipedia.org/wiki/Scientific_computing scientific computing] disciplines: [https://en.wikipedia.org/wiki/Computational_geometry computational geometry] and [https://en.wikipedia.org/wiki/Parallel_computing parallel computing]" [https://en.wikipedia.org/wiki/Parallel_mesh_generation]. The applications are listed (left to right) with respect to interactions of input form Computer-Aided Design to Sensor data like MRI, US, etc.]]

'''Broader Impact'''
The proposed activity will have a signiﬁcant and broader impact on research and education, well beyond the bounds of this speciﬁc project and its external collaborators. Other than the direct impact to parallel unstructured mesh generation and runtime systems for supercomputers like Blue Waters, this project will beneﬁt the performance evaluation community. This project will also promote the understanding on CerebroVascular Disease (CVD, or stroke), Image Guided Surgery for Caner, and DBS for Parkisons disease, all of which are leading natural death causes in the US.

'''Intellectual Merit'''
The Intellectual Merit of this project is summarized in terms of three novel contributions: (1) an understanding of the complex computation, communication, and memory utilization patterns in a telescopic superposition of multiple parallel mesh generation codes, (2) new abstractions to efficiently throttle concurrency and to implement race-to-halt execution models at multiple levels of granularity, while orchestrating intra-and inter-layer memory management, data movement, and load balancing to improve energy-efficiency of memory and interconnect components, and (3) a novel approach to leveraging off-the-shelf and state-of-the-art low core meshing codes for highly parallel simulations.

== Extreme-Scale Adaptive/Anisotropic Parallel Mesh Generation ==
<gallery heights=300px widths=300px mode="slideshow">
File:telescopic_and_meshes.jpg
File:CDT3D slide rocket nacelle.jpg ‎
File:Dlr airbus.jpg
File:Scitech 2019 delta wing.jpg
</gallery>

<gallery heights=200px widths=200px mode="packed-hover" caption="CDT3D Meshes using speculative execution model at the chip level">
File:CDT3D.jpg|130px|CDT3D Mesh
File:Rocket_Mesh.jpg|130px|DLR-F6 Airbus
File:Output_grid.jpg|130px|Rocket with Engine Mesh
|- style="text-align: center;"
||||'''CDT3D Meshes using speculative execution model at the chip level'''
</gallery>

=== Overview ===
Finite Element Mesh Generation is a critical component for many (bio-) engineering and science applications. The goal of this project is to deliver a novel Telescopic framework for highly scalable and energy efficient codes. Domain-and application-specific knowledge and run-time system support are combined to improve accuracy of FE computations.

CRTCLab addresses in depth two pillars: (i) high performance computing software runtime systems and (ii) mesh generation & adaptation based on error-metrics for NASA’s Computational Fluid Dynamics (CFD) Vision 2030. Specifically, CRTC’s focus is on the following three objectives: (1) design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation using state-of-the-art functionality supported by algorithms from AFLR and CRTC’s telescopic approach for parallel mesh generation , (2) develop error-based metrics adaptive anisotropic FE mesh generation methods and (3) design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretizations with localizable error estimates. This is a joined project with NASA/LaRC and MIT. 

<li style="display: inline-block;">
[[File: SubGroup1.jpg|thumb|right|500px| '''CRTC's sub-group for Extreme Scale Parallel Mesh Generation & Adaptation, from left to right: Nikos Chrisochoides, Kevin Garner, Dana Hammond (former TM from NASA/LaRC), Christos Tsolakis and Polykarpos Thomadakis. ''']]
We have assembled a team of established leaders (see [https://cepm.cs.odu.edu/People#External_Collaborators External Collaborators]) that are currently developing state-of-the-art work on mesh generation and adaptivity issues relevant to NASA’s CFD 2030 Vision and will broadly impact end-user productivity of users throughout DoD and NASA.

'''Objectives'''
:# Design a multi-layered algorithmic and software framework for 3D tetrahedral anisotropic parallel mesh generation methods using state-of-the-art functionality supported by methods implemented in AFLR and CRTC’s telescopic approach for parallel mesh generation.
:# Development of error-based metrics to drive an anisotropic adaptive process
:# Design a power-aware parallel runtime software system for extreme-scale adaptive CFD computations including: (i) mesh generation & adaptation, and (ii) consistent error-based metrics for adaptation of any CFD discretization with localizable error estimates.
</li>

== Real-time Adaptive Mesh Generation in Medical Image Computing ==

This application is using adaptive (isotropic) mesh refinement to improve (in real-time in the context of image-guided neurosurgery) the accuracy of 3D Physics-Based Deformable Registration of intra-operative MRI with pre-operative MRI.

=== Overview ===
<li style="display: inline-block;">
[[File: CT imaging.jpg|thumb|right|500px| '''Deep Brain Stimulation – CT Imaging obtained from Neurosurgical Section of SE PADRECC McGuire VAMC (with VCU)''']]
:* Develop adaptive deformable registration method for Deep Brain Stimulation (DBS). DBS is an effective palliative therapy for patients suffering from Essential Tremor, Parkinson’s disease, and other neurological movement disorders. Modern DBS surgery makes use of stereotactic systems and image guidance to accurately place electrode leads, as well as intra-operative imaging to surveil the location of the leads and guide the surgery. The effectiveness of DBS is directly correlated with the accuracy of DBS electrode lead placement, with more accurate electrode placement leading to better clinical outcomes. This project is a joined with VCU, VA. 
</li>

<li style="display: inline-block;">
[[File: Health-care.jpg|thumb|left|500px| '''APBNRR ''']]
:* Develop a new Adaptive Physics-Based Non-Rigid Registration (APBNRR) method developed for image guided brain tumor resection. Existing evaluation indicates that the the registration accuracy and performance of APBNRR are proved sufficient to be of clinical value in the operating room. Malignant gliomas are the most common primary and metastatic brain tumors. Treatment typically includes surgical removal followed by radiotherapy or chemotherapy. Total resection is difficult to achieve during tumor removal because of the infiltrative nature of gliomas and because brain tumors are often embedded in critical functional brain tissue. In image-guided neurosurgery, co-registered pre-operative anatomical, functional, and diffusion tensor imaging can be used to facilitate a maximally safe resection of brain tumors in certain areas of the brain. However, because the brain can deform significantly during surgery, particularly in the presence of tumor resection, non-rigid registration of the preoperative image data to the patient is required. This is a joined project with Harvard Medical School, MA. 
</li>

<li style="display: inline-block;">
[[File: ExtremeScale.jpg|thumb|right|500px| '''Images created by the I2M tool using the CBC3D Method''']]
:* Test five new available (in the clinic) neurovascular devices that have not been compared before through angiographic washout analysis and accurate micro CT-based tessellation (i.e., mesh or grid reconstruction) and CFD analysis in idealized sidewall aneurysm models. The CFD simulations and angiographic washout analysis will be performed on flow diverter implantations in actual patient geometries. There is a need for robust and easy-to-use Image-to-Mesh (I2M) conversion methods and software that streamline the discretization of such complex vascular geometries with fine structures with the CFD modeling process. In this project we will focus on the development and validation of fully functional I2M conversion algorithms and software (i.e., tools) for complex vascular geometries with fine structures like stents. This is a joined project with Medical School at Stony Brook University, NY. 
</li>

File:CRTC ProteinCavities D614G 8.png

2020-07-07T18:50:13Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 8.png

File:CRTC ProteinCavities D614G 7.png

2020-07-07T18:50:03Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 7.png

File:CRTC ProteinCavities D614G 5.png

2020-07-07T18:49:52Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 5.png

File:CRTC ProteinCavities D614G 4.png

2020-07-07T18:49:45Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 4.png

File:CRTC ProteinCavities D614G 3.png

2020-07-07T18:49:39Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 3.png

File:CRTC ProteinCavities D614G 2.png

2020-07-07T18:49:32Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 2.png

File:CRTC ProteinCavities D614G 1.png

2020-07-07T18:49:23Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 1.png

File:CRTC ProteinCavities D614G 8.png

2020-07-07T18:43:38Z

Angelos: Angelos uploaded a new version of File:CRTC ProteinCavities D614G 8.png