Health Research and Policy


DATE: October 24, 2013
TIME: 1:15 - 3:00 pm
LOCATION: Medical School Office Building, Rm x303
TITLE: Statistical Analysis of Rare Conformational Changes in Molecular Dynamics Simulations
SPEAKER: Sergio Bacallado
Stein Fellow, Department of Statistics, Stanford University

Molecular dynamics simulations are computer experiments which use a physics-based potential energy function to replicate the dynamics of molecules in solution. They offer insights into the ensemble of 3-dimensional structures that proteins and other biological macromolecules experience, and may be useful to understand phenomena as diverse as enzymatic function, transport, and binding. The main difficulty lies in the fact in that simulations are tremendously expensive and scale poorly with the size of the system. A related challenge of a statistical nature is to assess the quality of estimates for long-timescale behavior derived from a large number of short simulations performed in a massively parallel fashion.

In this talk, I will review several results on this problem. I will describe (i) bounds on the quality of Markov models on a discretization of phase space, (ii) a Bayesian nonparametric analysis of reversible Markov models on a discretization of phase space, and (iii) a new, nonparametric estimator for transition probabilities in the full phase space.

Suggested readings:
Sergio A. Bacallado, Stefano Favaro, Lorenzo Trippa. Bayesian nonparametric analysis of reversible Markov chains. The Annals of Statistics, 41, 2, pp. 870-896, 2013.

Marco Sarich, Frank No, Christof Schuette. On the approximation quality of Markov state models. Multiscale Modeling and Simulation. Society for Industrial and Applied Mathematics, , 8:4 (1983) 1154-1177, 2010.

Footer Links: