CROSSTCR: a curated resource for TCR cross-reactivity and off-target risk assessment in precision cancer immunotherapy

T cell receptor (TCR) cross-reactivity, the ability of one TCR to recognize multiple peptide–MHC (pMHC) complexes, is essential for immune adaptability but poses major safety risks for TCR-engineered cancer therapies, as molecular mimicry between tumor and self-peptides has led to severe off-target toxicities. We developed CROSSTCR, a manually curated, cross-reactivity-focused database integrating experimentally validated evidence of TCR cross-reactivity. Bidirectional TCR–antigen relationships are annotated, and data reliability is ensured through rule-based filtering, expert curation, and multi-level verification using sequence alignment, ESM-2 embedding similarity, and structural comparison. The version 1.3 dataset analyzed in this study contains >3400 curated TCR–pMHC interactions across ~1400 cross-reactive groups. Peptides recognized by the same TCR showed closer ESM-2 embedding proximity than matched controls, and structure-resolved subsets displayed convergent peptide backbone conformations and side-chain presentation. A retrospective analysis of the MAGE-A3/Titin mimicry event recapitulated the known molecular basis of clinically observed off-target cardiotoxicity. The web platform ( https://i.uestc.edu.cn/CROSSTCR ) provides interactive visualization and downloadable datasets. To our knowledge, CROSSTCR is the first public resource specifically designed to organize experimentally supported TCR cross-reactivity relationships, supporting molecular mimicry assessment and evidence-guided safety evaluation for TCR-based cancer immunotherapies.