On Scoring and Binding Affinity Changes Prediction in Protein-Protein Interactions

Abstract

The aim of this thesis is to promote the understanding of protein-protein interactions (PPIs) by gaining insights into their 3D structures at atomic level and their binding thermodynamics. It focuses on two main challenges in this field, namely the scoring problem, i.e. the identification of near-native conformations from a large pool of docking models, and the ∆∆G prediction problem, i.e. the prediction of binding affinity changes upon mutations in protein-protein complexes. For this, two machine learning-based computational methods, iScore and iSEE, were developed for scoring and ∆∆G prediction, respectively, and a curated ∆∆G database named DACUM was built, which provides information about the experimental methods used for measuring binding affinity changes.