Improvement of protein-peptide interaction prediction performance using 3 molecular docking software

Abstract

Proteins play a crucial role in many biological processes, where their interaction with other proteins are integral. Abnormal protein-protein interactions (PPIs) have been linked to various diseases including cancer, and thus targeting PPIs holds promise for drug development. However, experimental confirmation of the peculiarities of PPIs is challenging due to their dynamic and transient nature. As a complement to experimental technologies, multiple computational molecular docking (MD) methods have been developed to predict the structures of protein-protein complexes and their dynamics, still requiring further improvements in several issues. Here, I report an improved MD method, namely three-software docking (3SD), by employing three popular protein-peptide docking software (CABS-dock, HPEPDOCK, and HADDOCK) in combination to ensure constant quality for most targets. I validated our 3SD performance in known PPI references. I also enhanced MD performance in proteins having intrinsically disordered regions (IDRs) by applying the modified 3SD strategy, the three-software docking after removing random coiled IDR (3SD-RR), to the known PPI references. At the end, I applied 3SD-RR to the AlphaFold2-predicted receptors, yielding an efficient prediction of PPI poses with high relevance to the experimental data regardless of the presence of IDRs or the availability of receptor structures. This thesis provides an improved solution to the challenges in studying PPIs through computational docking and has the potential to contribute to PPIs-targeted drug discovery.