TCRDiff: Conditional diffusion model enables antigen-specific T-cell receptor generation

Antigen-specific recognition by T-cell receptors (TCRs) is crucial for adaptive immune responses. De novo generation of the complementarity-determining region 3 (CDR3) loops of TCRs offers a computational alternative to laborious experimental screening for antigen-specific TCR discovery. However, current approaches are constrained by weak conditional guidance and limited flexibility. Here, we introduce TCRDiff, a generative diffusion framework for designing antigen-specific TCRs conditioned on peptide-MHC (pMHC) targets and germline-encoded TCR variable genes. Pre-trained on large-scale T-cell repertoires and TCR-pMHC recognition pairs, TCRDiff generates native-like CDR3$\alpha\beta$ sequences via a denoising diffusion process. Incorporating interface geometry features yielded TCR-pMHC complexes with greater structural plausibility than models relying solely on sequence-based diffusion or structure-based modeling. As a proof of concept, we employed TCRDiff to design candidate TCRs against a tumor-associated antigen. These results establish a powerful computational framework and support the development of TCR-based immunotherapy.