Kinetics-Optimized Enhanced Sampling Using Mean First Passage Times

Molecular dynamics simulations have become essential in many areas of atomistic modelling from drug discovery to materials science. They provide critical atomic-level insights into key dynamical events experiments cannot easily capture. However, their impact often falls short as the timescales of the important processes are inaccessible using standard molecular dynamics. Enhanced sampling methods provided avenues to access such crucial rare events, for example key slow conformational changes of biomolecules. However, the bias in enhanced sampling simulations is rarely optimized, and even if they are, the optimization criteria is based on the thermodynamics or Hamiltonian of the system, but do not directly consider molecular kinetics. Here, we introduce a novel enhanced sampling algorithm that adaptively optimizes the bias based on the kinetics of the system for the first time. We identify the optimal bias that minimizes a key physical observable, the mean first passage time (MFPT) from a starting state to a target state. Our algorithm makes use of the relation between biased and unbiased kinetics obtained from discretized Markov state models (MSMs), as established in the dynamic histogram analysis method (DHAM). We demonstrate the applicability of the method for different 1D and 2D analytical potential-based model examples, NaCl dissociation in explicit water, and phosphate unbinding in Ras GTPase. Our algorithm has excellent performance compared with state-of-art enhanced sampling methods in terms of the timescales required to reach the final state in the benchmarking systems. Our findings provide a novel, kinetics-driven enhanced sampling strategy, signatured by a targeted approach to facilitate mapping rare events, with the potential for breakthrough applications in drug discovery and materials science.

Tiejun Wei, Balint Dudas, Edina Rosta · 2024