Contributing to the Understanding of Cell Membranes and Uremic Toxins by Combining Molecular Dynamics with Machine Learning and Statistical Modeling

Abstract

The calculation of thermodynamic and kinetic quantities related to biophysical phenomena is an integral part of understanding how biological systems function. Even the most basic interac-tions, such as ligand to protein residue, are critical pieces of information when interpreting complex biophysical processes. To this end, modeling of various systems and challenges, from detailed ki-netics calculations to specific interactions in nature, can be accomplished at an atomistic level using molecular dynamics (MD) simulations. MD provides an ideal opportunity to implement statistical modeling and, moreover, machine learning (ML) methods for deep system design and results anal-ysis due to the often large-scale high-dimensional data sets produced. In this dissertation, two dis-tinct biophysical settings are investigated. The first section studies model membranes and the ef-fects of small molecule partitioning, using Gaussian mixture models and enhanced sampling to identify potential changes in phase composition and solvation free energies. The second section begins by exploring uremic toxins bound to human serum albumin and provides long time-scale binding dynamics through Markov state models as well as detailed kinetic information regarding unbinding using deep learning to optimize unbinding coordinates. Finally, I model adsorbent pol-ymer materials for the capture of uremic toxins and design methods for screening a range of system configurations. This work highlights the wide range of biophysical challenges that can be rigorous-ly tackled by the combination of MD and statistical modeling/ML.