Coupling molecular dynamics and machine learning to accelerate the design of bioinspired materials

Abstract

Biological systems, including the human body, are directed by networks of exquisitely selective chemical interactions. Following the paradigm of bioinspired materials design, scientists and engineers aim to understand the rules of natural selectivity and to design materials that can achieve specific functions in complex biological environments without unintended consequences. Knowledge transfer from natural to synthetic design is limited by our understanding of specific and nonspecific interactions in nature at the molecular scale. Molecular dynamics (MD) simulations provide a means to study biophysical systems with atomistic detail. Machine learning (ML) techniques provide a means to synthesize meaningful insights from the deluge of high-dimensional data generated by MD. In this dissertation, I explore a range of important biophysical phenomena with MD and ML, yielding insights with unprecedented resolution and interpretability. Notably, I apply advanced simulation and unsupervised learning techniques to describe the molecular characteristics of nonfouling biomaterials and to guide the design of next generation renal replacement therapies for chronic kidney disease. This work exemplifies the potential for MD and ML in combination to accelerate bioinspired materials design.