Allmesberger-Riegler, L.; Frommelt, F.; L.Santini, B.; Superti-Furga, G.; Ferrada, E.; Sykacek, P.

Transmembrane (TM) protein-protein interactions (PPIs) are essential mediators of signal transduction, transport of solutes and communication, yet the biophysical features that characterize their diverse topologies remain poorly understood. To reduce this knowledge gap we use the human solute carrier (SLC) interactome to study whether physicochemical features of PPI interfaces encode information about their TM topology (i.e., their position and arrangement with respect to the membrane). To this end we predicted structures for 2,055 experimentally validated PPIs using AlphaFold v3.0, performed molecular dynamics simulations and annotated the PPI interfaces by TM coverage. As a result every interface is characterized by 63 physicochemical, structural and energy features. A reproducible machine learning workflow allows us to study the interdependence between these interface properties and interface topology. We found that amino acid composition and secondary structure contributed most to distinguishing soluble from fully membrane-embedded interfaces. Membrane-embedded interfaces contained a larger fraction of hydrophobic residues and -helices, while charged residues were depleted. For partially and full membrane-embedded interfaces amino acid composition and secondary structure get more similar and differences are increasingly observed in charge and energy related features. As particularly noteworthy we find that PPI interface characteristics vary with the number of annotated TM segments mainly through differences in proportions of secondary structure, charge, and flexibility. We hence conclude that PPI interface characteristics harbor substantial information about TM interface topology and provide a framework for the study and design of membrane protein interaction interfaces.