Fast and accurate committor estimation for kinetics simulations

TL;DR

The paper introduces a neural network-based, highly parallelizable algorithm for fast and accurate committor estimation, enabling efficient long-timescale biomolecular kinetics simulations.

Contribution

It presents a novel, efficient method combining neural networks and Milestoning for rapid committor calculation in biomolecular kinetics.

Findings

The method achieves high accuracy in committor estimation.

It significantly reduces computational cost for kinetics predictions.

Demonstrates robustness across complex biomolecular examples.

Abstract

Computing long-timescale kinetics of biomolecular processes remains a major challenge for atomistic simulations. A way out is to exploit local kinetic information to construct the global stationary flux across the reaction space. The committor serves as the optimal reaction coordinate for this purpose; however, its calculation is itself highly demanding. Here, we introduce a fast and accurate algorithm for committor estimation by leveraging highly parallelizable short trajectory simulations and analogue prediction. The resulting committor is represented via a neural network ansatz and subsequently coupled with the Milestoning method to predict the mean first passage time at very low computational cost. We demonstrate the robustness and efficiency of this committor-guided Milestoning (CoM) method through examples of increasing complexity.