Novel Methods for the Design and Analysis of Molecular Dynamics Simulations of Protein Systems

The application of atomistic molecular dynamics simulation to biological processes is a field of study with ever widening potential as computational power grows. However, the continual growth in the computational resources available to researchers provides a difficult challenge. As simulations become more complex and run for longer spans of time, our ability to analyze their behavior and understand the processes at work diminishes. The work detailed here seeks to address this problem by developing new computational tools, modifying existing tools, and applying machine learning methods to analyze and learn from molecular dynamics simulations. These methods include feature selection (Chapter 1), unsupervised clustering (Chapter 2), black-box optimization for simulation design (Chapter 3), proposing novel reaction coordinates for umbrella sampling (Chapter 4), and developing an A* search heuristic for allosteric pathway analysis (Chapter 5).