AI-guided transition path sampling of lipid flip-flop and membrane nanoporation.

Abstract

We study lipid translocation ("flip-flop") between the leaflets of planar lipid bilayers with artificial intelligence (AI) guided transition path sampling (TPS). Rare flip-flops compete with biological machineries that actively establish asymmetric lipid compositions. By initializing molecular dynamics simulations near transition states, AI for molecular mechanism discovery (AIMMD) captures lipid flip-flop without biasing the dynamics. Four distinct mechanisms of flip-flop emerge, as encoded in neural networks trained on the fly to predict the commitment probability (or "committor") for a lipid to proceed to one or the other leaflet. Whereas coarse-grained DMPC lipids "tunnel" through the hydrophobic bilayer, unaided by water, atomistic DMPC lipids cross the membrane through spontaneously formed water nanopores. In an atomistic plasma membrane mimetic, cholesterol tunnels unaided by water, whereas PLPC lipids exploit both transient water threads and nanodroplets to cross a locally thinned membrane, as seen also in an atomistic bilayer of DSPC lipids. Remarkably, in the high (~660) dimensional feature space of the deep neural networks in AIMMD, the reaction coordinate becomes effectively linear, in line with Cover's theorem and consistent with the idea of dominant reaction tubes.