Almost all biological processes are effectively mediated by protein-protein interactions (PPIs). However, our knowledge about PPIs lags behind our need for understanding cellular pathways, providing effective therapeutic intervention, and drug discoveries. Sequence coevolution approaches have recently led to a breakthrough in predicting monomer protein structures as well as protein interactions. Here we address, with EVComplex2, the ability to assess the likelihood of large-scale interaction prediction at residue resolution with a fast alignment concatenation method and a probabilistic score for the interaction of residue pairs. When assaying the E.coli cell envelope proteome, we predicted and resolved 467 PPIs including newly discovered 292 membrane protein interactions that are notoriously difficult to study experimentally. While EVComplex2 is a purely sequence-based method, it provides residue-residue restraints to construct structural models of protein-protein interactions. We provide tens of interaction models including the Flagellar Hook-Filament Junction, Tol/Pal System, and demonstrate the successful application of the method to the eukaryotic human spliceosome complex.
Predictions are available on https://marks.hms.harvard.edu/ecolicomple
EVcouplings Python Package is freely available on https://github.com/debbiemarkslab/EVcouplings