How is molecular dynamics being used in life sciences? Davide Branduardi

Transcription

1 How is molecular dynamics being used in life sciences? Davide Branduardi

2 Exploring molecular processes with MD Drug discovery and design Protein-protein interactions Protein-DNA interactions Protein folding Liang Z et al., PLOSOne 2011 Stranges et al., 2011 PNAS van Dijk et al., Nucl. Acid Res

3 What is molecular dynamics? Given the 3D coordinates of the system (PDB), evolve it through Newton s equations of motion while keeping in standard conditions It produces a trajectory (the so-called ensemble ) initial positions t=0 Initialize velocities for T=300 K t=0 Calculate the force t=tn How many iterations we make in a 1-ns simulation? one timestep is 1 fs 1 fs = s 1 ns = 10-9 s 10-9 / = 106 (1 million) Finished Solve the equation of motion for all particles t=tn Update physical properties (P, T, etc..) t=tn Yes tn > tlast No Update timestep t=t+1

4 Why molecular dynamics? Get more from your crystal: time let us observe events happening at molecular time scales have full solvation, temperature and pressure effects Folding (tertiary struct.) Breathing motions Bond vibrations Folding (secondary struct) Ligand binding Protein aggregation Channel gating fs (10-15s) ps (10-12s) ns (10-9s) µs (10-6s) ms (10-3s) How long can we simulate? s Time

5 Desmond: state-of-the-art MD engine Developed by D.E. Shaw research to run on Anton (special purpose machine) Implements standard fixed-charges force field (Amber, Charmm, OPLS) and solvation models TIP3P, SPC/E, TIP4P. No covalent bond breaking. Schrödinger enhancements: Support for OPLS3 force-field (improved protein and nucleic acids force field + enhancement on ligand parametrization) Integration with Maestro modeling environment Wealth of analysis tools for protein and protein/ligand interactions Extends FEP and enhanced sampling modules Watermap

6 Desmond molecular dynamics on CPU Single/double precision Scalable, can run on many CPUs Best cross-node performance is with fast interconnect (+Linux VM) *caveat: Desmond requires a Linux environment to run, analyses can be performed also on the other supported architectures

7 Desmond molecular dynamics on GPU Desmond on GPU only NVIDIA cards supported (Driver v or newer) single precision only cannot run across multiple cards Gamer s card (GTX series) Professional card (Tesla series) ~3-4 GB of memory (less than 400K atom systems) *Standard support does not cover consumer-level GPU cards such as GeForce GTX cards. ~12 GB in memory

8 Desmond on CPU vs GPU DHFR system (23K atoms): full solvated system with empirical force field 350 Performance (ns/day) K20 K40 K80 Due to HW and efforts from DE Shaw research in software performance improvements (also in memory footprint)

9 Preparing a typical MD simulation workflow Prepare System read in coordinates from PDB, or generate them yourself add missing atoms, add amino and carboxy terminals Add missing loops Chose alternate loc. solvate neutralize add salt Minimize reconcile observed structure with force fields used ( T = 0 K) Heating and Equilibration Ensure system is stable at target temperature Simulate Simulate under desired conditions Analyze Collect data Evaluate observables (macroscopic level properties) relate to single molecule experiments

10 Simulation setup: 2GMX example

11 Simulation setup: 2GMX example For this specific exercise select chain A, create a new entry from it, and process only this one

12 Protein reliability report

13 Protein preparation wizard Go! You might want to enable these if there are missing loops. Not needed here. You might want to disable this option to preserve the crystal water molecules

14 Protein preparation wizard Remove sulfates: used fro crystallization but not present in physiological conditions (there is no problem in parametrizing them though in OPLS3)

15 Protein preparation wizard Step A: protonati on & Asn, Gln flips Step B: restrained optimizatio n Note: every step produces a new entry in the project table so you have a backtrace of the modifications on your system

16 Protein preparation wizard Local minimization already improves significantly the structure

17 Build the simulation box * If you do not see the Applications panel you need to go to Task->Applications view

18 Build the simulation box * *requires to access a Linux host in schrodinger.host

19 Build the simulation box The box is build by wrapping a preequilibrated and repeated water box around the protein. Overlapping waters are removed Ions are inserted to neutralize the charge FAQ: do I need to parametrize the ligand? No, OPLS2005 has some generic enough parameters to parametrize any ligand. The last OPLS3 forcefield includes more atom-specific high quality parameters. You can check if some of these are missing via Tools->Force Field Builder. If any is missing this can be parametrized via Macromodel constrained torsional scan, Jaguar DFT optimization and MP2 energy fitting (takes few hours: do it overnight)

20 Run the simulation Take the system which is on screen (must be the one you prepared with the System Builder ) 2. Optional: relax prior to production. Need always to do when you start with a PDB structure 3. MD expert section: ensembles but also positional restraints

21 Run the simulation Scale up to 10 ns for a therm Beware: small number = lot of data! A B Caveat emptor: MD runs can take days therefore it is not advisable to run straight from Maestro since the job control can be lost when Maestro is closed for example. Best practice: 1. Set up your run through the Molecular Dynamics interface 2. Set up your machine trough cog wheel -> Job Settings (where you can choose a machine that has GPU for example, as set in the schrodinger.host file) and press OK (not Run!) 3. Write the output files in your current working directory (you set this at the beginning with Project->Change Directory): cog wheel -> Write 4. Transfer the files produced (desmond_md_job_1.msj, desmond_md_job_1.cms and desmond_md_job_1.cfg) on the machine you selected for Job Settings 5. Use the command at the bottom of desmond_md_job_1.msj to launch it 6. Once finished, bring the output on your machine and load it in Maestro

22 Run the simulation Details of production run Initial position and forcefield A List of all the steps of minimization and therm/press to do Last line of msj contains the command you should use on the remote machine (note: you should remove # and replace C:\Users\DBranduardi\MyPrograms\Schro dinger2015-4_build14\utilities\multisim with $SCHRODINGER/utilities/multisim )

23 Relaxation steps What the directory looks like remotely Relaxation stage output: compacted for convenience Input Relaxation steps Production stage energy Production stage log Input Production stage checkpoint (to restart simulations) Overview about all the stages (relaxation included) Last conformation and trajectory containing dir (in pre 15-4 you need also a idx file) In green the files that you should copy back on your own machine for analysis in Maestro

24 Relevant analysis tools Simulation Quality Analysis: control the convergence of basic physical parameters Simulation Event Analysis: make an analysis of protein, ligand and observables that are relevant to this simulation (is ligand moving? Is a torsion changing?) Simulation Interaction Diagram: show how the ligand and protein interact during the simulation and check the protein properties

25 Simulation quality analysis First check to do in every simulation: temperature, potential energy and volume convergence Total energy is stable after initial transient Potential energy is stable after initial transient Temperature is constant Volume is converged

26 A look to the trajectory Import the *-out.cms file: this is linked to the trajectory directory (do not move the relative position of the two) -out.cms file contains the pointer to trajectory directory. Trajectory information is denoted by If you click you open trakectory panel Play trajectory interactively and center your reference system. You can also use VMD for that. Save frames for restarting, make movies

27 Simulation Events Analysis Drill down into the data by adding properties

28 Simulation Interactions Diagram Analyze property of ligand, protein and protein ligand Generate an automatic report in PDF format Contacts Contacts SSE RMSD L-Props. RMSF RMSF Rot. Bonds

29 Simulation Interactions Diagram

30 Simulation Interactions Diagram

31 Simulation Interactions Diagram

32 MD Modes with Desmond Standard Mode Free Energy Perturbation (FEP) Total free energy Absolute binding Absolute solvation Relative binding Protein mutation Ligand selectivity Protein-protein interactions Stability Enhanced sampling methods Metadynamics REMD REST Mixed solvent simulations Watermap

33 More bang for buck: enhanced sampling techniques Beyond standard MD Free energy via endpoint methods (via Free-Energy Perturbation) Free energy via enhanced sampling (REST, REMD, Metadynamics) Folding (tertiary struct.) Breathing motions Bond vibrations Folding (secondary struct) Protein aggregation Ligand binding Channel gating fs (10-15s) ps (10-12s) ns (10-9s) µs (10-6s) ms (10-3s) s Time Enhanced sampling methods can speed up your dynamics but require longer preliminary validation phases and some of them consist of an initial trial-and-error procedure (might be non high throughput)

34 Conclusions MD is fully integrated in the Schrödinger suite: protein preparation is similar to any other modelling task Building a suitable simulation box is just few clicks Desmond GPU implementation allows you to achieve great performance with just your workstation MD protocol includes an effective thermalization phase, completely transparent to the user Effective analysis tools allow to judge the quality of simulation and to trust your results providing valuable insights not available from the static structure

35 Acknowledgments Dima Lupyan Dan Robinson Jas Bachoo Woody Sherman Thank you for your attention!