AutoSIM: Redesigning and automating umbrella sampling for biomolecular conformational transitions

Abstract

Umbrella sampling (US) is a widely used biased simulation technique to generate projections of free energy surfaces (FES) of chemical and biomolecular processes along a reaction coordinate (RC). US results are sensitive to the choice and discretization of the RC along which bias is applied as well as simulation timescales. Here, we address these issues through a redefined US scheme based on recent statistical insights into sampling equilibration and convergence of RCs. This scheme is implemented in an algorithm—AutoSIM—that automates US runs based on a non-equilibrated molecular dynamics trajectory initiated in the native state basin. AutoSIM can generate FES sections associated with functional conformational transitions in biomolecules independent of a priori information on RCs or even the end state. Furthermore, AutoSIM allows for combining statistics from multiple US runs to systematically assess and improve the quality of FES projections along a given RC. We validate AutoSIM by comparing FES projections for alanine dipeptide and ubiquitin along the top principal components with those obtained from simulations without an external bias. Finally, we apply AutoSIM to capture the energetics for large-scale conformational transitions of HIV-1 protease between the open and closed forms of the enzyme.