Efficient sampling of protein conformational dynamics and prediction of mutation effects.

Abstract

Molecular dynamics (MD) simulation is a powerful tool enabling researchers to gain insight into biological processes at the atomic level. There have been many advancements in both hardware and software in the last decade to both accelerate MD simulations and increase their predictive accuracy; however, MD simulations are typically limited to the microsecond timescale, whereas biological motions can take seconds or longer. Because of this, it remains extremely challenging to restrain simulations using ensemble-averaged experimental observables. Among various approaches to elucidate the kinetics of molecular simulations, Markov State Models (MSMs) have proven their ability to extract both kinetic and thermodynamic properties of long-timescale motions using ensembles of shorter MD simulation trajectories. In this dissertation, we have implemented an MSM path-entropy method, based on the idea of maximum-caliber, to efficiently predict the changes in protein folding behavior upon mutation. Next, we explore the accuracy of different MSM estimators applied to trajectory data obtained by adaptive seeding, in which new rounds of short MD simulations are collected from states of interest, and propose a simple method to build accurate models by population re-weighting of the transition count matrix. Finally, we explore ways to reconcile simulated ensembles with Hydrogen/Deuterium exchange (HDX) protection measurements, by constructing multi-ensemble Markov State Models (MEMMs) from biased MD simulations, and reconciling these predictions against the experimental data using the BICePs (Bayesian Inference of Conformational Populations) algorithm. We apply this approach to model the native-state conformational ensemble of apomyoglobin at neutral pH.