Towards Real Time-Molecular Dynamics: Applications to Neutron Scattering Joseph E. Curtis* Mounir Tarek Douglas J. Tobias *NIST/University of Maryland Universite Henri Poincare, Nancy, France University of California, Irvine
Classical MD Simulations and Neutron Scattering MD F=-grad(U) -> { R(t), V(t) } Atomic detail responsible for NS Predict what cannot be measured Filtering tool to design experiments NS Complex environments for FF Readily calculable observables Overlapping time scale MD is becoming a commodity; ECCE/NWChem, VMD/NAMD, etc., but... Several “hurdles” remain for new users: (1)Yet another software program/language/OS to conquer (2)Setting up new systems in correct environments relevant for NS (3)MD parameters appropriate for NS (4)Analysis: data handling, write analysis codes, NS details (5)Limits of applicability of MD results (right and wrong & why?)
Goal: Lower the activation barrier to the generation trajectories from MD simulations to analyze neutron experiments Input: Coordinates... ‘black-box’ Output: NS Observables & Non-observables... Output: NS Observables & Non-observables (atomic and macroscopic) MD should become a transparent tool for the USER ATOM 1 N LYS RSA 127 ATOM 2 CA LYS RSA 128 ATOM 3 C LYS RSA 129 ATOM 4 O LYS RSA 130 ATOM 5 CB LYS RSA 131 ATOM 6 CG LYS RSA 132 ATOM 7 CD LYS RSA 133 ATOM 8 CE LYS RSA 134 ATOM 9 NZ LYS RSA 135 ATOM 10 H1 LYS RSA 136 ATOM 11 H2 LYS RSA 137 ATOM 12 H3 LYS RSA 138 ATOM 13 HA LYS RSA 139 ATOM 14 HB1 LYS RSA 140 ATOM 15 HB2 LYS RSA ~ more lines... MD specifically for NS
RT-MD STUCTURE HANDLER TOPOLOGY GENERATOR MD CODE ANALYSIS USER Structure & Connectivity Desired Observables Atomic Filters Convergence Criteria Open Source MD (NAMD, NWChem, Gromacs, PINY_MD) (Tcl/Tk) Wrappers Error Checking “Library” MD / NS Details Structures, FF MPI : Distributed Computing Manager Sampling Strategy Convergence Check Experimental Data OUTPUT Spectra Graphs/Data Images Summary
INPUT: STRUCTURE:{ R(0) } X-ray, NMR, NS, homology TOPOLOGY:{ U(q) } Connectivity Atomic details Inter-, Intra- U(q) ENVIRONMENT: Cluster Solution Crystal Powder Embedded systems Example: Immerse protein in a lipid MD: Observable Constraints/Restraints Prompt USER for parameters Automatic equilibration Production runs Distributed computing ANALYSIS: Data storage & reduction Experimental details R( ), I(q) MPI & distributed computing Convergence Post-run (re-)analysis PBC
SHORT-TIME WINDOWS RMSD MSF I(q,t) S(q, ) (q, ) G( ) Rho(z) LONG-TIME WINDOWS P2 S2 I(q) (SANS/SAXS) PRACTICAL EXPERIENCE Typical Runs: Equilibration: 0.1 to 1.0 ns Production: 0.5 to 20 ns 16 CPU cluster ~ 1 ns (1 day to a week) Data Sets: 10s of MB to 100s of GB Analysis Codes: Most NS calculations ~ minutes Some can take “days” --> MPI Spare Cycles: Multiple initial conditions, environments
Courtesy of Ryan Benz (UCI) CNBT computational team: L. Saiz (NIST) R. Benz, F. Castro-Roman, D. Tobias, S. Whilte (UCI) Membrane Structure: CNBT at NCNR S. White (UCI)
Membrane Structure by Direct Inversion The Problem: Experimental determination of atomic details of density profiles is too time consuming AND existing MD simulations are in error.
U(Z, ) = k z (Z - Z*) 2 + k ( - *) 2 Diagram by Stephen White Once validated, the idea is... On new/unknown membrane, measure one or two profiles (say, RC=CR’), use Z* and * ). Then, calculate membrane properties using restrained MD.
Useful? { R(0) } a model Hydration effects Dynamical averaging effects MPI Biomolecular Structure by MD-SAXS / MD-SANS Merzel and Smith PNAS 99 (8): 5378, 2002
Dynamics: NS and MD Experiment: estimate mean-squared displacement from elastic intensity via Debye-Waller factor: I(0) = exp(–Q 2 ) Simulation: calculate resolution-broadened S(Q,E) as FT of I(Q,t)R(t), where R(t) is the FT of the instrument resolution function
Dynamics of N and MG states in solution: neutron scattering vs. MD MD gives excellent representation of dynamics of native -lactalbumin MD qualitatively reproduces enhanced broadening (i.e. additional motion) in MG QENS shows more broadening in MG vs. N state because MG sample contains substantial population of more highly unfolded states MD provides atomic details necessary to generate more robust analytical models MD vs. QENS on disk chopper TOF instrument at NIST ( ~ 100 ps) Tarek et al. Chemical Physics 292, , 2003 NativeMolten globule
Model Free Approach and NMR Relaxation Data n 2 H NMR on a calmodulin-peptide complex with partially deuterated methyl groups (48 of 79). Lee & Wand, Nature 411, , n Methyl group dynamics quantified by generalized order parameters obtained by fitting relaxation data using Lipari & Szabo “model free” approach
Order parameter extrapolation Neutron data (Doster et al.) SolutionDehydrated Powder
Summary Tools exist for “black box” MD Flexible framework; new MD and analysis code Mature MD techniques & analysis code for NS Structure and dynamics (day(s) & GBs) Next Steps? Pick a builder Carefully evaluate MD codes for NS Carefully evaluate MD codes for computing infrastructure Link computer scientists and MD/NS experts