Efficient sampling of the conformational space of biomolecular systems is a challenging problem with direct relevance to drug discovery research. Among the numerous enhanced sampling techniques that seek to address this challenge, the weighted ensemble (WE) method is attractive insofar as it is unbiased regardless of the chosen hyperparameters. The variance versus cost of WE-based estimates, however, depends strongly on the choices of hyperparameters. These choices include the way in which the conformational space is subdivided into “bins” and the allocation of trajectory walkers across these bins. Aristoff and Zuckerman recently proposed techniques for optimizing the choices of binning and walker allocation, given an observable whose steady-state average is to be computed. Here we describe our efforts to automate their hyperparameter optimization strategy within a workflow for performing weighted ensemble molecular dynamics simulations on the Orion cloud computing platform. We discuss preliminary applications of the optimized WE approach to estimate reaction rates in molecular systems of varying complexity. By developing a better understanding of how to apply the optimized WE approach to different pharmaceutically relevant systems, we hope to reduce the guesswork required to achieve efficient sampling in our large-scale WE simulations.

Exploring a hyperparameter optimization strategy for weighted ensemble molecular dynamics in the cloud

Fall 2022 abstracts, posters, and presentations will launch in mid-September.

ACS Fall 2022

The official archive and repository for all ACS Meetings & Expositions abstracts and select posters or presentations.