CCV, ICES, University of Texas at Austin
Biomoleculer Visualization and Computations at CCV
Angstrom
Vinay K Siddavanahalli
Center for Computational Visualization
Institute of Computational and Engineering Sciences
Department of Computer Sciences,
University of Texas at Austin
Ribosome, ribbon imposter rendering
CCV, ICES, University of Texas at Austin
Application domain
• Visualization– Volume, Imposter and Isosurface models– Grid / client server rover based.– Compression based, and hardware accelerated algorithms
• Animation– Flexible models– Volumetric video compression and interactive rendering
• Bioinformatics– Quantitative– Qualitative– Topological
• Protein docking– Compressed format, with flexibility information
CCV, ICES, University of Texas at Austin
Model creation – from the PDB database
PDB files
Electrondensity
Electrostaticpotential
SES, SCS
Volume + isocontours
Volume + isocontours
Linear, higher order meshes
Imposter rendering
Volume rendering of Rice dwarf virus
Volume rendering of hemoglobin
CCV, ICES, University of Texas at Austin
2D Image Processing
Reconstruction
3D Image Processing/Modeling
Particle PickingParticle Picking
ClassificationClassification
Cryo-EMImages
ParticleImages
EstimatedOrientations
Alignment
& Averaging Alignment
& Averaging
Groups ofParticles
3D ElectronDensity Map
Refi
nem
en
t
Adaptive ContrastEnhancement
Adaptive ContrastEnhancement
AdaptiveFilteringAdaptiveFiltering
2D/3D Image Enhancement and Correction
CTF CorrectionCTF Correction
3D Image Segmentation
3D Image Segmentation
Asymmetric Units
Medial Axis Extraction
Medial Axis Extraction
Helices/SheetsDetection
Helices/SheetsDetection Shape Matching Shape Matching
FeatureExtraction Feature
Extraction
SecondaryStructures
Pseudo-atomicStructure
s
Gaussian Blurring
Gaussian Blurring
ProteinData Bank
with other information
Orientation
Determination
Orientation
Determination
ParticleAverages
Reconstruction from 2D to 3D
Reconstruction from 2D to 3D
Model creation – from imaging datasets
CCV, ICES, University of Texas at Austin
Data structure
• We use a combined hierarchical Volumetric, Surface and Bond-level structural representation.
• Compressed data is used for time varying volume rendering and storage. We are also working on using it for other visualization algorithms including isosurface extraction.
• There are two distinct pipelines we follow to produce our datasets– From the PDB. ( from which we receive bond level information )
– Imaging data sets of large biomolecules.
CCV, ICES, University of Texas at Austin
Protein specific data structureGroups of proteins
Protein a
Chain 1
Protein p
Secondary structure 1 Secondary structure s
Residue 1
Chain c
Residue r
Atom list
• Since we use a hierarchical data structure for the bond-level domain, proteins can be represented naturally.• Bond information, like connectivity and torsion angles along the backbone are also maintained for flexibility modeling and visualization• Level of detail function computations and rendering is facilitated in this model.• It is extensible; level can be added, removed easily and each level uses arrays than lists to enable fast array rendering.•Each level is the same data structure, could just subclass to add more to it.
CCV, ICES, University of Texas at Austin
Multiresolution images
HemoglobinResidues
Secondary structures
Backbone chains
CCV, ICES, University of Texas at Austin
Volumetric visualization• Volumes are generated either through Gaussian blurring ( to produce
density maps ) or through APBS to obtain electrostatic potential maps.– Use texture based hardware rendering.– A hierarchical data structure on the bond level allows us to generate a
multiresolution model of the volumetric fields.• The multiresolution format is useful for level of detail rendering and adaptive protein
docking.
• The volume data structure we use is a RAWV format. It is a header which contains a description of the data set, followed by the grid positioned voxel vector values.
• Internal structure is a 3d grid and a colormap structure.
CCV, ICES, University of Texas at Austin
Internal Data Storage, Access
• DataManager has different DataSet Arrays• Each dataType is associated with API,
renderer, widgets• The DataManager has a generic API with
calls including load, delete, render etc.• The DataSet implements general IO
functions, including capabilities, presence of expected properties etc.
CCV, ICES, University of Texas at Austin
Bond level rendering• Large surface rendering can be prohibitive for interactive
rendering.• We use an imposter based model to render the ball and stick
model. Only one rectangle per primitive ( like sphere or cylinder ) is required. Depth and normal mapping yields true high quality surfaces.
• Further speed up is achieved through our hierarchical model representation.
Interactive rendering of the 1.2 million atom microtubule using the imposter model on PCs with NVIDIA
programmable graphics cards
CCV, ICES, University of Texas at Austin
Mesh generation• Adaptive Volume Meshes are required for obtaining adaptive potential fields.• Here, a simple listing of primitives is used as the file format rather than vrml
or stl etc. Internally, surface meshes are stored and handled as isosurfaces
94847 vertices and 497327 tetrahedrons
The active site groove is inside the red box. Adaptive meshes are generated in order to keep the accuracy of the groove, and reduce the number of elements at the same time.
AcetylCholinesterase (2573)
CCV, ICES, University of Texas at Austin
Flexibility modeling
• Bond angles representation for hierarchical modeling of flexibility.
• Volumetric video compression scheme for interactive rendering of 3d time varying data
Time varying volumetric videoShowing the hemoglobin action.Data by Dr.David Goodsell
CCV, ICES, University of Texas at Austin
Compression based Computational Visualization
• We use compression for the following:– Storing , streaming large datasets, including
isocontours and volumes and time varying volumes.
– Represent functions of proteins in a hierarchical manner to:
• Render interactively and use Level of Detail algorithms
• Perform protein docking
CCV, ICES, University of Texas at Austin
Linear Hierarchal BasisTC:571
Haar WaveletsTC:571
CCV, ICES, University of Texas at Austin
Rate Distortion (2EZP)
010203040506070
0.070.0
80.1
00.1
50.2
80.4
50.8
11.0
71.2
51.5
01.7
22.8
5
bits/voxel
PS
NR
(d
B)
Linear HBHarr
Rate Distortion (Hemoglobin)
0
10
20
30
40
50
60
0.08
0.10
0.14
0.20
0.29
0.45
1.06
1.72
2.30
2.85
3.00
bits/voxel
PS
NR
(d
B)
Linear HBHarr
CCV, ICES, University of Texas at Austin
Interrogative Visualization
• Query with a PDB file for additional information– Potential fields– Curvature calculations– Topological information– Fast isosurface mesh extraction
• Quantitative information– We have developed the contour spectrum, which we can use to obtain
quantitative information like volume, surface and gradient information.– This supplements visualization for our understanding of the data sets
• Time varying volumes– Track time varying quantitative changes, like volumes of components. This helps
to understand the change in properties of the biomolecule as it changes over time.
Mean curvature of 1a06
CCV, ICES, University of Texas at Austin
APIs
• Many libraries like isocontouring , volume rendering are easy to interface to. ( inputs, outputs easy to define, understand )
• Imposter based rendering uses slightly different information format, but very similar to the hierarchical GroupOfAtoms data structure.
• Volume , topological, quantitative queries can be made again as calls to libraries.
CCV, ICES, University of Texas at Austin
Resources
• CCV software can be downloaded from http://ccvweb.csres.utexas.edu
• We are recently working on grid enabled scientific visualization. – Collaborators include Steve Cutchin (SDSC),
Erik Engquist (SDSC), Art Olson (TSRI), Michel Sanner (TSRI)
CCV, ICES, University of Texas at Austin
Acknowledgements
• CCV – Dr C Bajaj– Julio Castrillon– Peter Djeu– SK Vinay– Zeyun Yu– Bong-Soo Sohn– Young-In Shin– Sangmin Park– Yongjie (Jessica) Zhang – Greg Johnson– Zaiqing Xu– KL Chandrasekhar– Qiu Wu– Jasun Sun– Anthony Thane– Shashank Khandelwal
• Computational resources– CCV/ICES/UT
– NPACI/SDSC
• Sponsors– NSF
– UT/MDACC/Whitaker
– NPACI/NSF
– DOE-LLNL/Sandia