COMPREHENSIVE MARKOV STATE MODELS FOR ASSESSING AND IMPROVING THE ACCURACY OF PROTEIN FOLDING SIMULATIONS

Abstract

Computational studies have become an essential tool in biochemistry, providing detailed insight into biological systems alongside experimental studies. Molecular simulation can predict protein conformational dynamics and the impact of mutations, enabling rapid and low-cost investigation of potential therapeutic targets and better understanding of biological systems. Molecular dynamics (MD) is a computational method able to model ensembles of biomolecular conformations in solution by simulating atomic motion at high temporal resolution. The principle limitation of MD is the ability to collect sufficient data for equilibrium sampling. However, with the progression of high-performance computing (HPC) clusters and distributed computing platforms, timescales previously inaccessible to MD can be reached and relevant protein parameters can be extracted using modeling. From these simulations, Markov state models (MSMs) are used extract system-relevant kinetic and thermodynamic information. An MSM represents a series of memoryless, probabilistic transitions between discrete states in a kinetically meaningful way. The obtained information is used to understand the relationships between relevant protein conformations, thus enabling a comprehensive understanding of the modelled system in a human-readable format. Recent advancements in model scoring and hyper-parameterization moved MSM construction away from anecdotal, case-by-case basis to a highly systematic approach that focuses on optimization and validity. Thus, modern MSMs are employed to investigate protein properties, and predict experimental observables using system-representative ensembles of conformations. Additionally, a comprehensive MSM can be combined with sparse experimental data to generate an improved interpretation of the system. My work focuses on performing all-atom massively-parallel MD simulation using the Folding@home distributed computing platform in order to build comprehensive MSMs that are used in improving simulation accuracy and protein design. This work results in the development of an unbiased framework for MSM building that is used to lend insight into simulation parameters, extract novel system behavior and enable clear comprehension of a target function, such as impact of mutations or emphasis of rare events.