Yin, RuiAntibodies and T cell receptors (TCRs) are crucial for the immune system's ability to recognize and combat pathogens and cancer cells. High resolution structures of antibody-antigen complexes and TCR-peptide-MHC (TCR-pMHC) complexes provide key insights into their targeting. This knowledge has enabled the structure-based design of vaccines against viruses and pathogens, and therapeutics against cancer, immunological disorders, and viral infection. However, the vast diversity of the immune repertoire, along with limited resources and time constraints, makes experimentally determining the structures of most antibody-antigen and TCR-pMHC interactions challenging. To support these experimental efforts, computational approaches have been developed to model the structures of these protein-protein interactions. Despite decades of development, an accurate predictive understanding of the structural basis of antibody and TCR targeting remains a challenge. Recently, deep learning algorithms have shown major promise in the field of molecular modeling, due to their ability to analyze and learn complex non-linear features underlying molecular systems. For my research, I harnessed the power of deep learning tools toward predictive modeling of antibody and TCR recognition. First, I examined the structural and physiochemical features underlying antibody-antigen recognition for antibodies that interact with the SARS-CoV-2 receptor-binding domain (RBD). Then, as a critical step toward the development of highly accurate modeling tools, I conducted a thorough benchmarking of the state-of-the-art deep learning algorithm, AlphaFold, in modeling protein-protein complexes. Focusing on antibody-antigen complexes, I identified critical areas where AlphaFold's modeling capabilities could be enhanced. Next, I developed improvements of AlphaFold to perform accurate modeling of TCR-pMHC complexes, leading to the TCRmodel2 algorithm, which is available to the community as a public web server. This was followed by an effort to explore the use of increased sampling to improve AlphaFold success, which generated near-native predictions for approximately half of antibody-antigen test cases and nearly all TCR-pMHC test cases. These advances in modeling accuracy constitute a leap forward in our predictive understanding of immune recognition and can serve as a step toward successful design of more effective vaccines and therapeutics.enDissertationBiology