Surface-Aware Generative Design of Selective Inhibitors for Protein–Membrane Adsorption

Abstract

Peripheral membrane proteins (PMPs) bind membrane surfaces through interfacial binding sites (IBSs) that are critical for stable membrane association and proper orientation of catalytic active sites. Dysregulated PMP–membrane interactions can enhance catalytic activity and contribute to diseases including cancer, but many PMPs remain difficult therapeutic targets because their active sites lack well-defined pockets or become accessible only upon membrane binding. In this work, we developed a multistage computational framework that integrates MaSIF-PMP, a surface-centered geometric deep learning model for IBS prediction, with a generative diffusion model to design protein binders that target PMP IBSs and inhibit membrane adsorption. Candidate binders were ranked using MaSIF-PMP prediction scores and geometric descriptors, then validated by molecular dynamics and enhanced sampling simulations. Applied to three disease-relevant PMPs, the framework identified selective binders that disrupted membrane association while minimizing off-target membrane interactions, providing a potential strategy for designing PMP membrane-adsorption inhibitors.