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

Abstract

2 Phase separation in mixed lipid systems has been extensively studied both experimentally and 3 theoretically because of its biological importance. A detailed description of such complex systems 4 undoubtedly requires novel mathematical frameworks that are capable to decompose and 5 categorize the evolution of thousands if not millions of lipids involved in the phenomenon. The 6 interpretation and analysis of Molecular Dynamics (MD) simulations representing temporal and 7 spatial changes in such systems is still a challenging task. Here, we present a new unsupervised 8 machine learning approach based on Nonnegative Matrix Factorization, called NMFk, that 9 successfully extracts physically meaningful features from neighborhood profiles derived from 10 coarse-grained MD simulations of ternary lipid mixture. Our results demonstrate that leveraging 11 NMFk can (a) determine the role of different lipid molecules in phase separation, (b) characterize 12 the formation of nano-domains of lipids, (c) determine the timescales of interest and (d) extract 13 physically meaningful features that uniquely describe the phase separation with broad 14 implications. 15 3 16 Author Summary17 Cell membranes contain mixtures of chemically diverse lipid molecules, dynamically structured, 18 that play a key role for cell survival. Under different conditions, lipids in the membrane can 19 segregate into liquid-ordered and liquid-disordered domains. It is well-known that such domains 20 play a critical role in cellular homeostasis. Unfortunately, the lack of molecular details hampers 21 the direct interpretation of the available experimental data for such lipid segregation dynamics.22 Molecular Dynamics (MD) simulations can potentially fill the gap between theory and 23 experiments, but this task requires rigorous methods for proper analysis and description of the MD 24 generated trajectories. Here, we present a new unsupervised machine learning analysis based on 25 Nonnegative Matrix Factorization that extracts physically meaningful features from pre-processed 26 datasets generated by MD simulations. We simulate and analyze phase separation in a well-studied 27 ternary lipid mixture. Our results demonstrate that via the proposed machine learning we can: (a) 28 determine the role of different lipid molecules in the phase separation, (b) characterize the 29 formation of nano-domains in lipid bilayers, (c) determine the timescales of the formations of these 30 nano-domains and (d) extract physically meaningful features that uniquely describe the dynamics 31 of the phase separation 33 34 Cell membranes contain mixtures of different lipid types, dynamically arranged, that play a key 35 role in various mechanisms responsible for cell survival [1]. In the past, membranes were thought 36 to be homogenous systems, however, new data suggests that under different stimulus, the lipids 37 can segregate [2,3] into detergent-resistant domains commonly called "rafts". These domains are 38 highly dynamic, varying in size and composition [4,5]. Important...

Phase Separation in Liposomes Determined by Ergosterol and Classified Using Machine Learning

“…Therefore, when making a diagnosis, it will be possible to provide suggestions for improving the system. Previous studies demonstrated that lipid bilayer membranes (Walter et al, 2020) and lipid mixtures (Löpez et al, 2019) contribute to understanding how phase separation occurred using machine learning systems. These aforementioned studies, along with the present research, aim to understand phase separation on lipid membranes using machining learning systems.…”