Development of transferable and chemically accurate coarse-grained models for lipid bilayers using a dynamic particle swarm optimization algorithm

Lipids are essential structural elements of cell membranes that have a number of biological functions, including cell compartmentalization, involvement in signaling cascades, and transport regulation. The amphiphilic properties of lipids result in the formation of fascinating self-assemblies that are important for numerous biological processes. The chemical structures of lipids, including headgroup and acyl chain chemistry, determine the structural and mechanical properties of lipid bilayer assemblies. Thus, it is important to investigate these systems at the molecular level using efficient yet chemically detailed computational modeling methods, such as coarse-grained (CG) molecular dynamics (MD) simulations. Current shortcomings in lipid CG models include the loss of important physical information due to extensive coarse-graining as well as the incompatibility of models with polymeric systems, hindering the investigation of lipid-polymer conjugates for potential drug delivery applications. In order to tackle these shortcomings, we propose a new higher resolution, chargeless model for phosphocholine (PC) lipids that is transferable to different lipid families. We do this using a modified particle swarm optimization (PSO) algorithm that can dynamically drive the optimization process to regions of the search space which are most favorable based on on-the-fly estimations. Compared with optimization results obtained from the traditional PSO algorithm, we show that the convergence time for DPSO (dynamic-PSO) is 3-5 times faster which is essential for developing coarse grained models that are faster and cheaper. This reduced operation time is caused by increased convergence accuracy of the DPSO (through active local minima avoidance), removing the need for user input during the optimization process. The optimization is based on reproducing structural (area per lipid, membrane thickness) and mechanical (bending rigidity) properties of PC lipid membranes obtained from sophisticated techniques, including small-angle X-ray scattering and neutron spin echo (NSE). The model we present is universally applicable with fair agreement with experimental results. Furthermore we present a structurally accurate model of cholesterol that can be used to simulate more structurally and mechanically realistic bilayers for biomolecular applications.