Tcr-Translate: Conditional Generation of Real Antigen- Specific T-Cell Receptor Sequences

Abstract

The paradoxical nature of T-cell receptor (TCR) specificity, which requires both precise recognition and adequate coverage of antigenic peptide-MHCs (pMHCs), poses a fundamental challenge in immunology. Efforts at modeling this complex many-to-many mapping have focused on the detection of reactive TCR-pMHC pairs as a binary classification task, with little success on unseen epitopes. Here, we present TCR-TRANSLATE, a framework that adapts low-resource machine translation techniques including semi-synthetic data augmentation and multi-task objectives to generate target-conditioned CDR3β sequences for unseen input pMHCs. We examine twelve model variants derived from the BART and T5 model architectures on a target-rich validation set of well-studied antigens, find- ing an optimal model, TCRT5, that samples known and native-like CDR3β se- quences for unseen epitopes. Our findings highlight both the potential and lim- itations of sequence-to-sequence modeling in rapidly generating antigen-specific TCR repertoires, emphasizing the need for experimental validation to bridge the gaps between predictions, metrics, and functional capacity.