Structure-based prediction of the peptide sequence space recognized by natural and synthetic PDZ domains.

摘要: Protein-protein recognition, frequently mediated by members of large families interaction domains, is one the cornerstones biological function. Here, we present a computational, structure-based method to predict sequence space peptides recognized PDZ largest recognition proteins. As test set, use considerable amount recent phage display data that describe peptide preferences for 169 naturally occurring and engineered domains. For both wild-type domains single point mutants, find 70-80% most observed amino acids are predicted within top five ranked acids. Phage identified different from those in original crystal structure. Notably, about half these cases, our algorithm correctly captures preferences, indicating it can mutations increase binding affinity relative starting We also computationally recapitulate specificity changes upon mutation, key successful forward design protein-protein interface specificity. Across all evaluated sets, incorporation backbone sampling improves accuracy substantially, irrespective using or NMR structure as conformation. Finally, report prediction several acid blind tests DREAM4 domain challenge. Because foundational methods developed here based, results suggest approach be more generally applied redesign other interfaces have structural information but lack data.