par Li, Dong 
Président du jury Gilis, Dimitri
Promoteur Pucci, Fabrizio
Co-Promoteur Rooman, Marianne
Publication Non publié, 2025-08-29
Président du jury Gilis, Dimitri
Promoteur Pucci, Fabrizio
Co-Promoteur Rooman, Marianne
Publication Non publié, 2025-08-29
Thèse de doctorat
| Résumé : | The recognition and removal of foreign pathogens by the immune system rely on molecular interactions between lymphocyte receptors and antigenic molecules. Specifically, B-cell receptors or antibodies bind directly to antigens, while T-cell receptors (TCR) recognize antigenic peptides presented by major histocompatibility complexes (pMHC). These natural mechanisms have inspired the development of therapeutic agents such as monoclonal antibodies, vaccines and T-cell-based therapies. Their advancement greatly depends on a detailed understanding of the molecular basis of antibody-antigen and TCR-pMHC binding. Despite the significant computational and experimental efforts devoted to these issues over the past decades, these immune recognition mechanisms are still far from being fully understood. Following this line of research, this PhD thesis presents a series of computational methods for analyzing and predicting the structures of antibody-antigen and TCR-pMHC complexes. First, we developed ImaPEp, a convolutional neural network-based pipeline that predicts binding probabilities between antibody paratopes and antigen epitopes. By encoding each antibody-antigen interface as a 2D image that captures physicochemical properties and spatial layout, ImaPEp achieves fast and accurate performance in cross-validation and in large-scale antibody-antigen pair screening tasks. Second, we constructed AbAgym, a dataset including single amino acid mutational effects on antibody-antigen interactions, extracted from deep mutational scanning experiments. First, we used it to carefully analyze antibody-antigen interfaces and identified key hotspot residues that contribute to complex stabilization. Additionally, we benchmarked several computational prediction methods on this dataset, finding that physics-based approaches outperform others. The AbAgym resource is not only valuable for interface analysis, but also enables both the benchmarking and training of computational predictors aimed at forecasting the effects of mutations on antibody-antigen interfaces. Third, we introduced PInteract, a computational tool for detecting π interactions within and between chains in a protein/DNA/RNA 3D (complex) structure. These interactions play a critical role in both antibody-antigen and TCR-pMHC recognition. By incorporating both distance- and angle-based criteria, PInteract efficiently identifies various classes of π interactions. It offers a user-friendly command-line interface and can be a useful resource for gaining insight into the role of these interaction types in protein function, stability, and binding. Finally, we conducted a structural analysis of TCR-pMHC complexes to elucidate key molecular determinants underlying TCR recognition, in comparison with antibody-antigen recognition. This analysis highlights specific physicochemical and topological features that correlate with TCR-pMHC binding affinity and specificity. Although the question of why TCRs are MHC-restricted remains elusive, our findings provide new insights into the structural constraints that may shape this binding. Together, the computational tools developed in this work and their applications advance the study of immune recognition and contribute to the rational design of immune-targeting therapeutics, with potential impact across biomedical research, immunotherapy, and vaccine development. |
