Automatic Generation of Physiologically Relevant Lipid Bilayers

Lawrence Livermore National Laboratory
2020–21

A lipid bilayer separates a cell from its surroundings and determines what substances, such as therapeutic drugs, may enter the cell. Because in situ research on cell membranes is difficult and drug development is costly,  research on the behavior of proteins embedded in lipid bilayers is often done with molecular dynamics simulations. This project has studied the output of a continuum model recently developed at LLNL and found that the lipid concentrations therein can be described by a multivariate gaussian model. This model powers an application, GRuMPy (Generative Membranes in Python), that allows researchers to construct, simulate, and analyze realistic membranes starting from incomplete specifications. As part of this process, we developed and performed two validation tests to ensure that there was appropriate probability density in the generative model over compositions that are reasonable in the continuum model. We also fit and tested an asymmetry model that allows the application to take percent compositions as input and output full membranes with appropriate proportions of lipids in the inner and outer leaflets. In addition, we provide a tool that allows researchers to train and deploy generative models based on new or updated continuum model output.

Given that we generate membranes from a model, and that GRuMPy provides the ability to simulate the compositions that it generates, we also sought to validate the generated compositions as reasonable. Validation tests done on membrane simulations of size 5 nm to 30 nm using compositions generated by the application show that the membranes generated have simulated properties identical to membranes in the continuum model. Compositions generated by the application show no significant deviation from the continuum model in area per lipid, bilayer thickness, area compressibility, or order parameter. In effect, with the assumption that the continuum model data is composed of reasonable membrane compositions, GRuMPy is able to generate physiologically realistic membranes.

Advisor(s): Peter N. Saeta.

Team: Emma Frances Cuddy ’21, Rebecca Fei Qin ’21, Rakia Segev ’21, Eric McLaughlin Weiner ’21, and Rachel Emily Cohen ’21.