Supervised Learning on Markov States as a Method for Gaining Insight into Protein Allostery

Abstract

Understanding complex allosteric communication in biophysical systems has long been a challenging problem in computational chemistry. Recent advances in Markov state modeling has made the problem more tractable but the automatic detection of important degrees of freedom remains a significant challenge. Using applied supervised learning methods combined with conformational space discretization via Markov Modeling, we are now able to predict important degrees of freedom in molecular dynamics (MD) simulations. These methodological advances allow us to quickly ascertain potentially important degrees of freedom leading to a better understanding of allostery in large enzymes.