UC Riverside

The ability to simulate biological phenomena across a wide range of scales has been steadily improving for decades. The atom-level approach known as Molecular Dynamics (MD) has proven remarkably useful for rationally engineering proteins, studying peptide binding, and detailing protein folding and unfolding pathways. Though the utility of MD is difficult to overstate, its simulations are expensive, relying on integrating equations of motion across exceedingly small timescales. This creates a large discrepancy between the timescales accessible to MD, on the order of microseconds, and those of biomolecular phenomena, on the order of milliseconds to seconds. The work presented here serves as a deep dive into studying patterns in MD data, formulating approaches that extrapolate phenomena from short simulations and construct a platform by which new samples may be efficiently generated. The first half of this dissertation focuses on analyzing MD data from complicated molecular machinery, beginning with a deep dive into a CRISPR Associated Transposon (CAST) system called Cascade-TniQ. A set of approaches is implemented to unveil subtle binding sites that influence conformational changes and to demonstrate synchronous motions, yielding a detailed mechanistic basis for the thermodynamics of the DNA-binding process. A second deep dive into CRISPR/Cas13a inspired the development of a method studying cross-talk efficiency known as the Signal-to-Noise Ratio, detailing the dynamic allostery responsible for activating the complex upon DNA binding. The second half introduces a neural network, SurfNet, informed by these deep dives into Cascade-TniQ and CRISPR/Cas13a. SurfNet learns a function of atomic coordinates known as a collective variable (CV) to guide MD-based enhanced sampling methods. These methods efficiently explore the space of CV outputs but require good CVs, an unsolved problem for most proteins. SurfNet was validated on the prototypical test systems of Alanine Dipeptide and Chignolin using On-the-fly Probability Enhanced Sampling (OPES). Ultimately, the work presented here is meant to assist the sample generation and analysis journeys that computational biophysicists embark on to learn as much as we can about the brilliant molecular machines known as proteins.