computational biophysics - bioinfosummer 2012 (jason roberts)

33
Case Study: Atomistic Molecular Dynamics Simulation of Poliovirus Jason A. Roberts, Senior Medical Scientist, National Enterovirus Reference Laboratory, WHO Regional Poliomyelitis Reference Laboratory, VIDRL Applied Sciences, RMIT University, Australia.

Upload: australian-bioinformatics-network

Post on 04-Jul-2015

377 views

Category:

Documents


2 download

DESCRIPTION

Case Study: Atomistic Molecular Dynamics Simulations of Poliovirus

TRANSCRIPT

Page 1: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Case Study: Atomistic Molecular Dynamics Simulation of Poliovirus

Jason A. Roberts, Senior Medical Scientist,National Enterovirus Reference Laboratory, WHO Regional Poliomyelitis Reference Laboratory, VIDRLApplied Sciences, RMIT University, Australia.

Page 2: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

�Animation –� Creates the illusion of movement

� Simulation –� Tool for testing a hypothesis

Page 3: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Study of a molecule(s) structure and function using computational methods such as:

� Semi-empirical quantum mechanics

� Atomistic Molecular Dynamics

� Coarse-Grained Molecular Dynamics

� Hybrid methods

� Ultimately these are just models!

Page 4: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

0.01

0.10

1.00

10.00

100.00

1,000.00

10,000.00

100,000.00

1992 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012

TF

OL

PS

Year

Supercomputer Peak Speed (RMax)

vs Time (years)

Rmax (TFLOPS)

First 1µs simulation of enzyme folding (BPTI)10,000 atoms (4 months)

Desktop + GPU

Cray Titan

17.59 PFLOPS

iPhone

3.3 million atom poliovirus or

rhinovirus 1µs simulation

(4 months) Eg. Avoca VLSCI 2012

BlueGene/Q using 25% of RMax

4U Server + 8GPU

Page 5: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� BlueGene/Q (VLSCI@25%) = 8.9 days� 65,536 threads (~172 TFLOPS)

� BlueGene/P = 3 months� 2048 cores (~7 TFLOPS)

� 4u GPU Server (CUDA) = 3 months� 32cores + 8GPUs (~10.5 TFLOPS)

� Desktop GPU (CUDA) = 1 year� 12core + 2GPU (~2.2 TFLOPS)

� Modern Laptop = 20 years� 4core i7 (~0.15 TFLOPS)

� iPhone, Android = 300 years� (~0.01 TFLOPS)

Page 6: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 7: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 8: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Side-chain behaviour

� Hydrophobicity

� Variations in pH conditions

� Side-chain to side-chain interactionseg. S-S bonds in Cysteine

� Salt Bridges

� Long-range electrostatics (eg. PME)

Page 9: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Nucleic Magnetic Resonance� Small structures, very high resolution.

� Provides some conformational state data

� X-ray crystallography� High resolution

� Requires production of a crystal

� Cryogenic Electron Microscopy� Biologically representative, relatively low resolution

Page 10: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Based on existing data attained by traditional methods eg:� Nuclear Magentic Resonance (<1.0 Å)

� X-ray crystallography (0.5-10 Å) most ~2.0 Å

� Cryo-EM (~4-20 Å)

� Able to simulate various biological environments

� Place a static structure in a simulated biological environment

Page 11: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

X-Ray Crystal

NMR

CryoEM

Page 12: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 13: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

PubMed search

Polio – 22,922 (c1879 onwards)

Poliovirus – 14,381 (c1951 onwards)

Picornavirus – 30,708 (c1945 onwards)

RCSB records

– Poliovirus (56)

- 41 X-Ray

- 8 EM

- 7 NMR

– Picornaviridae (207)

- 180 X-Ray

- 19 EM

- 8 NMR

Page 14: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Optimal match selected manually from RCSB-Protein Data Bank and phylogenetic analysis

� Models created using SWISS-Model with template 1HXS (2.2 angstrom resolution, most complete chain information)

� Matrix data used from template to recreate full capsid using VMD multiseq for 3D alignment and mon2poly script to generate new chains

� Custom parameter files generated using SwissParam website

Page 15: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Figure derived from Fields Virology 6th edition and Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 16: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Quantum mechanics� Realistic throughput of ~100 atoms

� Ab-initio modelling (eg. protein folding)� μs – ms timeframe with multiple conformations = very

unreliable and computationally expensive.

� Homology Modelling

% Amino Acid Identity

Unreliable OK – Careful checking Good - Reliable Excellent

30% 60% 90% 100%

Page 17: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Virus reconstruction

� 241 protein chains (52,812 Amino Acids)

� 120 lipids (60 covalent bonded to N-terminus VP4)

� 7.5 kb RNA genome

� Total virus model >1 million atoms

� Simulation size 4.2 million atoms (Cubic PBC)

� More than 3 million water atoms and ions!

� ~15-40 million CPU hours to simulate 1μs!

Page 18: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 19: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Rhombic dodecahedron 3.3 Million atoms

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 20: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 21: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� Topology file generation� Ion placement� Data transfer and analysis

� Files for creation, simulation and analysis

▪ 0.1μs simulation time = 50GB of files (0.1ns steps)

� Simple calculations (RMSD/F) for protein coat

▪ 10ns trajectory data at 0.1ns time points

▪ 52,800 amino acids x 100 intervals

▪ 5.28 million data pointsExcel spread sheets need not apply

Page 22: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 23: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 24: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 25: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

A. Pentamer B. Empty Capsid C. Full Virus

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 26: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

A. Empty Capsid B. Full Virus

Figure derived from Roberts, JA ., et al DOI 10.1016/j.jmgm.2012.06.009

Page 27: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 28: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 29: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 30: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 31: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)
Page 32: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� exaFLOP (1018) computing� Better scaling via architecture and software

improvements� petaFLOP = >10ns/day full virus� exaFLOP ~= >10μs/day full virus� Simulations approaching billions of atoms� Newer force fields to emulate quantum

mechanics� Increased integration of MD in research

Page 33: Computational Biophysics - BioInfoSummer 2012 (Jason Roberts)

� A/Prof. Bruce Thorley - VIDRL

� Dr. Mike Kuiper - VLSCI

� Dr. Andrew Hung - RMIT University

� Prof. Peter Smooker - RMIT University

� WHO Regional Poliomyelitis Reference Lab

� Tom Aitken

� Aishah Ibrahim

� Linda Hobday