Uncertainty quantification for Markov state models of biomolecules constructed using rare event acceleration techniques.

Abstract

Markov state models (MSMs) of biomolecular systems are often constructed using the molecular dynamics (MD) technique. Despite having very long MD trajectories, some states and pathways can be missing in the MD data, which may make the MSMs incomplete. Consequently, uncertainty quantification for the resulting MSM becomes important. Using deca-alanine as a prototype system, we demonstrate that rare-event acceleration techniques can be employed to greatly lower the MSM uncertainty with a high computational efficiency with the assumption that the rare-event acceleration technique is able to determine most pathways that are relevant to the dynamics. In particular, we explore applications of steered MD to construct MSMs. Upper and lower bounds for uncertainty in the resulting MSM are derived. Safeguards are built into our approach to handle scenarios where the rare-event acceleration technique is unable to discover some important pathways.