SLIDE 1
Molecular dynamics
- Molecular dynamics are stochastic
- Single trajectories are usually not representative
- Objective is to sample expectation values
High-throughput molecular dynamics simulation and Markov modeling - - PowerPoint PPT Presentation
High-throughput molecular dynamics simulation and Markov modeling - - PowerPoint PPT Presentation
High-throughput molecular dynamics simulation and Markov modeling Frank No (FU Berlin) frank.noe@fu-berlin.de Molecular dynamics Molecular dynamics are stochastic Single trajectories are usually not representative
SLIDE 2
SLIDE 3
Conformation Dynamics / Markov models
ms - s ns - µs
Sampling Problem Reconciliation with Experiment Analysis Problem
hugedata sets
huge, complex datasets
SLIDE 4
100 ns / day / GPU*
e.g. Amber, AceMD, OpenMM
10 µs / day / Anton I Rate
50 K atom system (explicit solvent)
SLIDE 5
100 ns / day / GPU*
e.g. Amber, AceMD, OpenMM
10 µs / day / Anton I 100 GPUs 1 Anton I Throughput 10 µs / day 10 µs / day Rate 100 traj. of 100 ns / day 1 traj. of 10 µs / day
50 K atom system (explicit solvent)
SLIDE 6
100 ns / day / GPU*
e.g. Amber, AceMD, OpenMM
10 µs / day / Anton I Cost 100.000 USD 10.000.000 USD 100 GPUs 1 Anton I Throughput 10 µs / day 10 µs / day Rate 100 traj. of 100 ns / day 1 traj. of 10 µs / day
50 K atom system (explicit solvent)
SLIDE 7
System Size: 50 K atoms (100 ns/day/GTX780) Simulation lengths: 100 ns to 2 µs Total sampling data: 200 µs
Example for ligand binding: Trypsin – Benzamidine
pyemma.org
SLIDE 8
rare event not-so rare events bound unbound energy conformation
Good situation: diffuse landscape with intermediate steps
SLIDE 9
Schütte et al, J. Comput. Phys. 1999 de Groot et al, J. Mol. Biol. 2001 Swope, Pitera and Suits, JPCB 2004 Sriraman, Kevrikidis and Hummer, JPCB 2005 Schultheis et al, JCTC 2005 Singhal, Pande, JCP 2005 Chodera et al, JCP 2007 Noé et al, JCP 2007
Review book Early contributions Markov State Models Construction and Analysis
Dimension reduction Perez-Hernandez et al, JCP 2013 Estimation and Validation Prinz et al, JCP 2011 Computing kinetic experimental observables Noé et al, PNAS 2011 Computing transition pathways Noé et al, PNAS 2009
SLIDE 10
PyEMMA: Software for construction of Markov state models
www.pyemma.org
SLIDE 11
Trypsin apo conformation dynamics
pyemma.org
Plattner and Noé, Nature Comm. 6, 7653 (2015)
SLIDE 12
tunbind = 1.1±2 µs G = 6.5±0.8 kcal/mol tunbind = 0.03±0.001 µs G = 7.8±0.6 kcal/mol tunbind = 91±40 µs G = 3.6±0.6 kcal/mol Tunbind = 5.3±4 µs G = 4.3±0.7 kcal/mol tunbind = 394±141 µs G = 2.0±0.8 kcal/mol tunbind = 63±6 µs G = 0.03±0.2 kcal/mol Tunbind = 3.5±2 µs G = 4.9±0.7 kcal/mol
1* 1* 1* 1*
1 1 1
- r
1*
pij > 0.01 pij > 0.001 pij > 0.0001 pij < 0.0001
pyemma.org
Trypsin bound-state conformation dynamics
Plattner and Noé, Nature Comm. 6, 7653 (2015)
SLIDE 13
Overall binding / conformational kinetics
pyemma.org
Plattner and Noé, Nature Comm. 6, 7653 (2015)
SLIDE 14
Trypsin excited conformations found as ground-state conformations of other serine proteases
pyemma.org
Plattner and Noé, Nature Comm. 6, 7653 (2015)
SLIDE 15
Protein complexes Protein folding mechanisms
Faelber, …,Sadiq, Noé, Daumke Nature (2011)
0.52 0.52 0.52 7.4 2.57 5.35 5.35 Relaxed state Scission Constricted state +GTP
Noé et al, PNAS (2009) Sadiq, Noé, De Fabritiis, PNAS (2012)
Substrate binding to HIV protease
MSM: examples
Reubold et al, Nature (2015)
SLIDE 16
Current examples (in prep)
Rhodopsin (60 K atoms) Activation transition 1.5 ms Barnase-barstar (120K atoms) Protein-protein association 2 ms
SLIDE 17
”High performance” computing
Parallel nodes time
SLIDE 18
”High performance” computing
Parallel nodes time
SLIDE 19
Ensemble MD
Parallel nodes client pilot jobs time
SLIDE 20
Automatic / adaptive sampling machines
htmd.org copernicus-computing.org https://radicalensemblemd.readthedocs.org
SLIDE 21
Transition-based reweighting analysis method (TRAM)
SLIDE 22
Example: binding and folding of PMI to MDM2 fast (microseconds) slow (10-100 milliseconds) nanomolar binder
SLIDE 23
energy conformation rare event not-so rare events bound unbound
Bad situation: single high barriers (e.g. salt bridge)
SLIDE 24
energy conformation biased or generalized ensemble simulation (e.g. Replica-exchange, Metadynamics,...) direct MD bound unbound
Bad situation: single high barriers (e.g. salt bridge)
direct molecular dynamics biased or generalized ensemble simulation joint optimal estimate?
Wu, Mey, Rosta, Noé, JCP 141, 214106 (2014)
SLIDE 25
Example for an enhanced sampling method: Umbrella sampling + v(x) bi(x) . . .
Wu, Mey, Rosta, Noé, JCP 141, 214106 (2014)
SLIDE 26
Reweighting of equilibrium probabilities
equilibrium probabilites reweighting factor normalization constant Reweighted probabilities:
Wu, Mey, Rosta, Noé, JCP 141, 214106 (2014)
SLIDE 27
Transition-based Reweighting Analysis Method (dTRAM)
Estimation problem Constraints Reweighted probabilities:
Wu, Mey, Rosta, Noé, JCP 141, 214106 (2014)
pyemma.org
SLIDE 28
System Size: 50 K atoms (100 ns/day/GTX780) Trajectories: 500 x 1000 ns Total sampling data: 500 µs
Example: binding and folding of PMI to MDM2
SLIDE 29
Funding
Christof Schütte (FU Berlin) Eric Vanden-Eijnden (Courant Institut NY) Thomas Weikl (MPI Potsdam) Marcus Sauer, Sören Doose (Uni Würzburg)
Collaborations
Vijay Pande (Stanford) Katja, Faelber, Oliver Daumke (MDC) John Chodera (MSKCC NY) Gianni de Fabritiis (Barcelona)