Conserve Water: A Method for the Analysis of Solvent in Molecular Dynamics

Molecular dynamics with explicit solvent is favored for its ability to more correctly simulate aqueous biological processes and has become routine thanks to increasingly powerful computational resources. However, analysis techniques including Markov state models (MSMs) ignore solvent atoms and focus solely on solute coordinates despite solvent being implicated in myriad biological phenomena. We present a unified framework called "solvent-shells featurization" for including solvent degrees of freedom in analysis and show that this method produces better models. We apply this method to simulations of dewetting in the two-domain protein BphC to generate a predictive MSM and identify functional water molecules. Furthermore, the proposed methodology could be easily extended for building MSMs of any systems with indistinguishable components.