COMPUTATIONAL STUDIES OF PROTEIN-PROTEIN AND PROTEIN-ANTIBODY INTERACTIONS: IMPLICATION FOR MOLECULAR DESIGN.

High performance computing has opened the possibility to investigate complex systems by simulating their dynamics and study of equilibrium and non-equilibrium settings in realistic settings. Molecular Dynamics (MD) simulations have emerged as one of the privileged methods to disentangle the intricacies of biochemical systems but, despite the validity of Moore’s Law, the timescale of the events that can be simulated has an upper limit of the millisecond with tailor-made computers which is not enough to study some biologically relevant phenomena. Starting from these considerations, in this thesis, I have set out to develop and validate novel methods to predict the location of potentially interacting surfaces on proteins and to predict the impact of small molecules on the activation vs. the inhibition of proteins’ functional dynamic states. To this end, I have combined physico-chemical approaches to the study of protein dynamics and generate novel approaches that may overcome the current limitations of pure brute force MD simulations. In the first part of the thesis, I studied methods for the prediction of the residues involved in protein-protein interactions. I presented two different scores, one based on evolutionary information and one based on the energetics of the protein, on a dataset of crystal structures. Both scores have the capability to discriminate the interface region from the rest of the protein in a relevant fraction of cases. Moreover, a comparison of the scores efficacy on distinct protein classes highlights the importance of considering the biological function of the protein on the performance of the method used for the prediction of interface residues. In addition, the energetic method for interface residues prediction is used for the detection of antigenic epitopes on the spike protein of SARS-CoV-2. The regions predicted were confirmed against experimental complexes expanding our understanding of the molecular basis for interactions. In perspective, the acquired knowledge could be used for the design of novel vaccine candidates and diagnostic tools and to increase our readiness in the case of future epidemics. In the second part there, I focussed on the study of two allosteric systems. Firstly a method is presented for the integration of an ensemble docking protocol with a learning classifier for allosteric ligands of the protein Hsp90. The method reaches a good accuracy in classifying the activity of these ligands and this approach seems to reduce the dependency on the chemical similarity of the compounds used for the training. The method is tested on a limited dataset and further developments could be achieved in the future if the library of compounds is increased. In the end, I presented the initial analysis of an allosteric signal for integrinαvβ6 in complex with a pro-TGFβpeptide, with the use of molecular dynamics simulations. The data suggest that the presence of the peptide induces an increased rigidity of the legs of the structure, in particular for a specific domain.