Predicting membrane orientational states of RAS and RAS-RAF based on the lipid environment using deep learning.

Abstract

Cells utilize different active mechanisms to regulate cellular growth, differentiation, and survival. RAS and RAF proteins play a central role in these mechanisms, which are regulated by complex protein-lipid interactions. It is critical to identify molecular details by which local lipid environment modulates the behavior of these proteins to have detailed understanding of the mechanism underlying cancer formation. Here, we first employed deep learning (DL) to identify orientational states of RAS proteins and associated lipids using an extensive training data from a massive simulation campaign of 120,000 coarse-grained MD simulations each over 1 microsecond long, run using our MuMMI. Our results show that our DL model can predict protein states with an accuracy of over 80% based on the underlying lipid fingerprints and all the different lipid species contribute to the preferred RAS membrane orientational states.For the more complex RAS-RAF proteins, we employed DL (ResNet model) to explore the relationship between protein conformational structures and local lipids by predicting orientational states of RAS and RAS-RAF protein complexes based on the lipid densities around the protein domains. We obtained our training data from a recent RAS-RAF MuMMI campaign containing over 96 ms of aggregated CG simulation time. Our findings demonstrate that there is a significant correlation between the RAS and RAS-RAF orientational states and specific lipid densities around the protein domains. Our ResNet model can predict six protein states with an overall accuracy of around 80% from lipid density data. The findings of this work can assist in designing novel therapies targeting critical protein-lipid interactions involved in the mechanisms associated with cancer development. This work was performed under the auspices of the U.S. Department of Energy by Lawrence Livermore National Laboratory under Contract DE-AC5207NA27344. Release:LLNL-ABS-840775.