Development of new computational methods for the investigation of the molecular mechanisms underlying the interaction among proteins

Understanding protein binding mechanisms is fundamental to molecular biology, with significant implications for protein design and mapping the human interactome and complexome. Despite the important consequences for therapeutic and biotechnological applications, understanding the binding process and the stability of the resulting complexes is still a challenge. Predicting protein-protein interactions is difficult due to environmental factors and the complexity of the involved processes, which rely on geometric and chemical matches. The balance between accuracy and efficiency when choosing the features to consider is a central issue in computational techniques developed to predict complex formation. For example, poses rankings based on the scoring functions of docking servers frequently lacks precision. To better understand the mechanisms underlying protein complex formation, we developed a compact vector description of protein molecular surfaces based on orthogonal polynomial expansions. When applied to the evaluation of shape complementarity, our protocol allows for an indirect evaluation of van der Waals interactions, which are the predominant forces in protein binding at short distances; to improve the characterization of these van der Waals-dominated regions, we studied the role of electrostatic interactions, whose effect has been previously investigated mainly on long-rage. We found that binding interfaces exhibit a higher degree of electrostatic complementarity, defined as a spatial match between the signs of surface points facing each other, compared to random surface regions. We expanded the formalism to evaluate this feature: as for shape complementarity, this approach allowed us to quickly compare vectors describing surface regions without having to cal-static interactions not only facilitate the initial recognition and approach of proteins over long distances but also guide the reorientation of the interacting partners at shorter distances. In a second phase, complexes requiring more stable binding enhance their interlock through increased shape complementarity. Integrating these features and other physical and chemical characteristics with a neural network, CIRNet, allowed us to identify core interacting residue and improve docking algorithms by re-ranking proposed poses. CIRNet has demonstrated effectiveness across various types of protein complexes for three popular docking servers, reducing the average RMSD between the refined poses and the native state by up to 58%. culate the forces between all possible atoms pairings. We observed that electrostatic complementarity plays a key role in determining the stability of the binding: transient dimers show the highest elec- trostatic complementarity, while more stable complexes rely more heavily on shape complementarity. Interestingly, we noticed that shape complementarity is higher near the center of the interfaces, whereas electrostatic complementarity remains consistent across the entire binding region. These findings could suggest that electrostatic interactions not only facilitate the initial recognition and ap- proach of proteins over long distances but also guide the reorienta- tion of the interacting partners at shorter distances. In a second phase, complexes requiring more stable binding enhance their in- terlock through increased shape complementarity. Integrating these features and other physical and chemical characteristics with a neu- ral network, CIRNet, allowed us to identify core interacting residue and improve docking algorithms by re-ranking proposed poses. CIR- Net has demonstrated effectiveness across various types of protein complexes for three popular docking servers, reducing the average RMSD between the refined poses and the native state by up to 58%.