Unsupervised Machine Learning for Analysis of Coexisting Lipid Phases and Domain Growth in Biological Membranes

Abstract

Phase separation in mixed lipid systems has been extensively studied both experimentally and theoretically because of its biological importance. A detailed description of such complex systems undoubtedly requires novel mathematical frameworks that are capable to decompose and categorize the evolution of thousands if not millions of lipids involved in the phenomenon. The interpretation and analysis of Molecular Dynamics (MD) simulations representing temporal and spatial changes in such systems is still a challenging task. Here, we present a new unsupervised machine learning approach based on Nonnegative Matrix Factorization, called NMFk, that successfully extracts physically meaningful features from neighborhood profiles derived from coarse-grained MD simulations of ternary lipid mixture. Our results demonstrate that leveraging NMFk can (a) determine the role of different lipid molecules in phase separation, (b) characterize the formation of nano-domains of lipids, (c) determine the timescales of interest and (d) extract physically meaningful features that uniquely describe the phase separation with broad implications.