Markov State Models for Molecular Dynamics

The problem of protein folding is centered around understanding how the chain of amino acids comprising a protein folds into its stable, low-energy conformation. Because the laws of physics which govern the process are well-defined, protein folding can be simulated on computers.

However, doing so requires a tremendous amount of computation and generates a large amount of folding data.

This paper explains how to build Markov State Models to analyze protein folding data. As a form of probabilistic graph, Markov State Models can be used to process folding data generated from molecular dynamics data into a human-interpretable representation.

Under the Markov State Model framework, protein folding is viewed as a probabilistic process rather than a deterministic one, and uncertainty around estimates can be quantified and used to focus simulation efforts to reduce the number of computations needed by two orders of magnitude.