Improving the description of protein-protein association energy

Abstract

Proteins play a crucial role in virtually every biological process taking place within our cells. Most of the times, proteins do not participate in these processes alone but forming complexes of two or more proteins. Therefore, the study of protein-protein interactions (PPIs) and complex formation has become an important field of research in the last decades due to its scientific relevance and therapeutic interest. Protein docking is one of the several computational approaches that have been applied to study protein interactions over the last years. It aims to determine the three-dimensional structure of a protein complex based on the structure of its subunits. Although the field has experienced important advances in recent years, it faces significant challenges ahead. New strategies are necessary to overcome current sampling limitations and enhance the physico-chemical description of protein-protein association, understanding its intrinsic mechanisms and identifying the most relevant residues involved, i.e., hot-spot residues. This Ph.D. thesis has focused on developing new computational tools to address some of these challenges. We have developed pyDockLite, a simplified scoring function derived from pyDock, the docking scoring function developed within our lab, which is up to 10 times faster at comparable performance. The key element in pyDockLite development is the new distance-based desolvation term, which drastically reduces the computation time required to calculate the desolvation contribution to pyDock docking energy. Based on pyDockLite, we have developed a fast rigid-body minimization algorithm, which is very efficient when the complex subunits are in their bound conformation. To model backbone flexibility we have included normal modes in the minimization algorithm. This new feature improves the results, especially for the medium-flexible and flexible cases.

Most protein-protein docking protocols use scoring functions to evaluate docking poses and discriminate between good, i.e., near- native, and bad conformations. The implicit assumption is that the different energetic minima forming the docking energy landscape are represented by single docking poses which are scored individually. In this thesis, we have analyzed the concept that each energetic minima of the docking energy landscape can be formed by ensembles of docking orientations or conformations, and we have explored the consequences of scoring each minimum by such ensembles. We propose a novel ensemble-based description of the docking landscapes, integrating clustering, conformational sampling and consensus scoring, which improves docking performance. In some circumstances, we might want to have a more detailed description, at the level of residue or atoms, of the docking energy of the different states conforming the docking landscapes. We have developed a method to partition pyDock docking energy at the residue level. Interestingly, we will show how we can use this partitioned energy to identify energetically relevant residues in the binding process (hot-spots) and to estimate changes in binding affinity upon mutation to alanine, i.e., as an in-silico alanine scanning mutagenesis predictor. Regarding mutations to other residues, we have developed a new method to predict binding affinity changes upon mutation by combining MODELLER and pyDock. Results are in line with previous methods when tested on an external validation dataset. Finally, we have explored how to apply the knowledge and tools we have developed to other protein interactions such as those between proteins and RNA molecules. We present a new scoring function that combines FTDock score and pyDock electrostatics and van der Waals energy terms. This scoring function can be used to evaluate docking models of protein-RNA complexes. Our work indicates that protein-protein and protein-RNA interactions may have distinctive features that prevent the direct application of protein-protein scoring functions to protein-RNA docking studies