• Citer

Development of B-cell epitope prediction pipelines. Application to chronic lymphocytic leukemia and implications for disease origin.

par Cia Beriain, Gabriel
Président du jury Gilis, Dimitri
Promoteur Pucci, Fabrizio
Co-Promoteur Rooman, Marianne
Publication Non publié, 2024-04-26

Thèse de doctorat

• Accès en ligne
• Détails
• Contenu
• Statistiques

Résumé :

Antibodies are major molecules of the immune system raised in response to invading pathogens. They are characterised by their high binding affinity and specificity, as well as their vast antigen binding diversity. These characteristics have made antibodies attractive for the development of new therapeutics against diseases such as cancers, autoimmune disorders or infectious diseases. A better understanding of the characteristics of antibodies and their target antigens can therefore help in the development of new antibody-based therapeutics, as well as a better understanding of B-cell associated leukemias. This PhD thesis presents the results of multiple computational analyses performed on antibody and antigen data, and in particular in the context of SARS-CoV-2 and chronic lymphocytic leukemia (CLL).First, following a rigorous and independent benchmark, we showed that conformational B-cell epitope prediction methods are in general no better than random. We concluded that the B-cell epitope prediction task is probably ill-posed, and that it is likely that almost all protein surfaces can be bound by some antibody. Second, we performed an analysis of 83 neutralising antibodies with known 3D structures targeting the spike protein in the context of the SARS-CoV-2 pandemic. By comparing our computational analysis with experimental data, our results supported the existence of an immunodominant subset of antibodies, and indicated that they are characterized by an intermediate overlap with the binding interface of the ACE2 receptor, the entry point to the host cells, and, chiefly, by their ability to bind the closed conformation of the spike protein. On the basis of these results, we improved SpikePro, an in-house SARS-CoV-2 variant fitness prediction model based on the impact of the mutations on the spike protein's stability, binding affinity with ACE2 and binding affinity against an updated panel of immunodominant-only antibodies. SpikePro has been made freely and easily accessible to the scientific community through a webserver, which we hope will contribute to genomic surveillance efforts against newly emerging SARS-CoV-2 variants. Third, we have developed pyScoMotif, a fast and user-friendly tool for the search of identical and similar protein 3D structural motifs such as binding and catalytic sites. Through the use of an inverted index, pyScoMotif is able to return in a few seconds all the proteins with the given 3D structural motif in datasets containing hundreds of thousands of proteins. Finally, we have performed multiple analyses of antibody sequences from CLL patients, among which the evaluation of the structural similarity between patient's modelled 3D antibody structures as a potential prognostic biomarker, as well as the search of mimotopes similar to known B-cell receptor epitopes involved in homotypic interactions.