Development of an automatic tool to study the protein-membrane interface

Biological membranes are not passive scaffolds for membrane proteins but active regulators of their structure, dynamics, and function. In mammalian membranes, cholesterol plays a central role in modulating protein behaviour through both nonspecific effects on bilayer properties and specific interactions at the protein–membrane interface. Despite extensive experimental and computational efforts, the systematic identification of cholesterol binding sites remains challenging, as many experimental techniques provide static or ensemble-averaged views that obscure the dynamics and mechanistic details of lipid–protein interactions. Coarse-grained molecular dynamics (CG-MD) simulations offer a powerful means to overcome these limitations by enabling extended simulations of large protein–membrane systems over hundreds of microseconds, thereby capturing persistent and functionally relevant cholesterol interactions. However, the lack of standardized and automated workflows has hindered large-scale and comparative studies across membrane protein families. Here, we present HSDetector, an automated and scalable CG-MD–based protocol for the construction, simulation, and analysis of protein–membrane systems, designed to identify cholesterol interaction hot spots at the protein–membrane interface. The method integrates long-lived protein–cholesterol contact analysis with spatial density mapping to robustly distinguish persistent binding regions from transient encounters. We validate the protocol using two well-characterized systems, namely the cannabinoid 1 receptor and the serotonin transporter, and demonstrate its ability to recover experimentally observed cholesterol binding sites as well as previously unreported interaction regions. We further apply HSDetector to a representative benchmark of 56 class A G protein-coupled receptor structures, enabling a systematic, family-wide characterization of cholesterol interaction landscapes. This analysis reveals recurrent membrane-interface hot spots that are not reliably captured by sequence-based motifs, underscoring the importance of structure- and dynamics-driven approaches. Both the HSDetector software and the complete set of results are made publicly available as a Python package and an interactive online platform, providing a general framework for large-scale studies of lipid–protein interactions and supporting the discovery of membrane-facing allosteric sites relevant to drug design.