Project Summary T cell receptors (TCRs) are of increasing therapeutic interest due to the role of T cell mediated immune responses in conditions such as viral infection, cancer, cardiomyopathy, autoimmunity, and graft rejection. A T cell response begins when TCRs associate with peptide antigens presented by major histocompatibility complex (MHC) proteins on antigen-presenting cells. The formation of the TCR-peptide-MHC (TCR-pMHC) complex triggers an intracellular cascade that results in T cell activation and, for cytotoxic T cells, target cell killing. While a T cell response can be highly specific, TCRs cross-recognize multiple peptides. This feature, though biologically necessary, may cause off-target effects in therapeutic applications, as evidenced by tragic outcomes in clinical trials of T cell therapy. A challenge in developing safer T cell or TCR-based therapies thus lies in accurately predicting the cross-reactivity profile of a TCR - that is, the range and types of peptides to which it can and cannot respond. Current prediction methods are limited by a lack of high quality training data covering ranges of peptides, instead typically focusing on a single "cognate" peptide for each TCR, limiting the ability of prediction algorithms to generalize beyond what is already known. Various library-based or genetic screens have been developed, but these do not allow assessment of discrete peptides and prohibit control of relevant biologic variables. Others have tried positional scanning libraries (PSL), or X-scans, to probe the positional sensitivity of TCR recognition. While traditional PSLs overcome the limitations of other screens, they cannot probe the range of diversity needed to characterize a TCR’s cross-reactivity profile. I hypothesize that by systematically increasing the diversity of peptide libraries and integrating this data with advanced structural modeling and machine learning techniques, I can develop a more complete knowledge-base of the structural a