(169n) Surface-Centered Approach for Characterization and Prediction of Protein-Membrane Interactions

Peripheral membrane proteins (PMPs) are the membrane proteins that transiently bind to the membrane via surface interactions and participate in crucial cellular processes such as signaling cascades and lipid metabolism. Characterizing and predicting the membrane-binding interfaces of PMPs is important for developing novel therapies that disrupt PMP-membrane interactions or for rationally designing protein scaffolds to mediate interactions at target membranes. Predicting the membrane-binding interface is a challenging task, as the physicochemical properties of the interface are similar to those of nonbinding interfaces. The available structures of peripheral proteins are also obtained from their soluble states since stabilizing the membrane-bound state is difficult with contemporary experimental techniques. Thus, most mechanistic and energetic studies of PMPs have been performed using molecular dynamics simulations, but the time required to sample the whole protein-membrane interaction (PMI) process is tremendously large in case of all atomistic-level simulations. Therefore, developing methods to achieve fast predictions of PMP membrane-binding interfaces will broaden our understanding of the PMI process and can be employed for various purposes including the rational design of membrane-targeting drugs, biosensors with enhanced binding specificity and sensitivity, and antifouling coating for medical or marine industry.

This study seeks to understand and characterize the general surface properties of the interfaces between protein and membrane, then leverage this surface-centered understanding to predict PMP membrane-binding interfaces. To address this goal, we define numerical arrays, or fingerprints, that capture geometric and chemical features of the protein and membrane surfaces. These vectors are then used as input for a geometric deep learning model to predict the interaction between protein and membrane of interest and identify the specific regions of the PMP surface that correspond to the membrane-binding interface. To increase the accuracy of the approach and capture the flexibility associated with some membrane-binding domains, we perform molecular dynamics simulations to sample conformational dynamics as additional model input. We show the capability of this model to discriminate binding interfaces for a dataset consisting of over 1000 PMPs. This approach enables membrane-binding interface predictions and will further enable new mechanistic interrogation of binding modes to produce fundamental physical insight into the mechanisms that PMPs use to interact with membranes.