Consistent interpretation of molecular simulation kinetics using Markov state models biased with external information

TL;DR

This paper introduces a reweighting method within Markov state models to improve the consistency of molecular simulation kinetics with experimental or reference data, addressing model errors across multiple time scales.

Contribution

The authors develop a novel reweighting approach that integrates external kinetic constraints into Markov state models to enhance their accuracy and consistency.

Findings

Systematic improvement of time scale separation in peptide dynamics

Refined equilibrium properties through constrained rate adjustments

Identification of limitations in simulation methods via reweighting challenges

Abstract

Molecular simulations can provide microscopic insight into the physical and chemical driving forces of complex molecular processes. Despite continued advancement of simulation methodology, model errors may lead to inconsistencies between simulated and reference (e.g., from experiments or higher-level simulations) observables. To bound the microscopic information generated by computer simulations within reference measurements, we propose a method that reweights the microscopic transitions of the system to improve consistency with a set of coarse kinetic observables. The method employs the well-developed Markov state modeling framework to efficiently link microscopic dynamics with long-time scale constraints, thereby consistently addressing a wide range of time scales. To emphasize the robustness of the method, we consider two distinct coarse-grained models with significant kinetic inconsistencies. When applied to the simulated conformational dynamics of small peptides, the reweighting procedure systematically improves the time scale separation of the slowest processes. Additionally, constraining the forward and backward rates between metastable states leads to slight improvement of their relative stabilities and, thus, refined equilibrium properties of the resulting model. Finally, we find that difficulties in simultaneously describing both the simulated data and the provided constraints can help identify specific limitations of the underlying simulation approach.