Abstract B041: Predicting T-cell receptor specificity and phenotype using transfer learning models

Abstract T-Cell Receptors are at the heart of the immune system's response to antigens. Their vast heterogeneity, encoded via V(D)J recombination allows for recognition of a wide array of pathogens and cancer neoantigens. Several computational challenges arise when tailoring tratments to an individual's unique TCR repretoire, including mapping TCR sequences, especially CDR3b sequences to specific antigens and understanding and modeling the functional implications of TCR diversity, as well as cominging TCR data with other modalities of data, especially in single cell resolution. Here we propose a new algorithm for preditcting and modeling TCR phenotype and specificity to cancer antigens that utilizes a novel approach using transfer learning and large language models. We use antigen and epitope-specific single cell TCR sequencing data of putative cancer neoepitopes presented on MHC-I to extract and engineer CDR3 sequence-based features and train our model for TCR specificity prediction. Our work provides integration with single cell RNA-Seq modality in a comprehensive functional analysis. Citation Format: Marmar R Moussa. Predicting T-cell receptor specificity and phenotype using transfer learning models [abstract]. In: Proceedings of the AACR Special Conference in Cancer Research: Functional and Genomic Precision Medicine in Cancer: Different Perspectives, Common Goals; 2025 Mar 11-13; Boston, MA. Philadelphia (PA): AACR; Cancer Res 2025;85(5 Suppl):Abstract nr B041.