AlphaFold Ensemble\nCompetition Screens Enable Peptide\nBinder Design with Single-Residue Sensitivity

Understanding the relationship between the sequence and\nbinding\nenergy in peptide–protein interactions is an important challenge\nin chemical biology. A prominent example is ubiquitin interacting\nmotifs (UIMs), which are short peptide sequences that recognize ubiquitin\nand which bind individual ubiquitin proteins with a weak affinity.\nThough the sequence characteristics of UIMs are well understood, the\nrelationship between the sequence and ubiquitin binding affinity has\nnot yet been fully characterized. Herein, we study the first UIM of\nVps27 as a model system. Using an experimental alanine scan, we were\nable to rank the relative contribution of each hydrophobic residue\nof this UIM to ubiquitin binding. These results were correlated with\nAlphaFold displacement studies, in which AlphaFold is used to predict\nthe stronger binder by presenting a target protein with two potential\npeptide ligands. We demonstrate that by generating large numbers of\nmodels and using the consensus bound-state AlphaFold competition experiments\ncan be sensitive to single-residue variations. We furthermore show\nthat to fully recapitulate the binding trends observed for ubiquitin,\nit is necessary to screen AlphaFold models that incorporate a “decoy”\nbinding site to prevent the displaced peptide from interfering with\nthe actual binding site. Overall, it is shown that AlphaFold can be\nused as a powerful tool for peptide binder design and that when large\nensembles of models are used, AlphaFold predictions can be sensitive\nto very small energetic changes arising from single-residue alterations\nto a binder.