Assistant Professor University of Oklahoma Norman, Oklahoma, United States
Disclosure(s):
Marmar Moussa, PhD: No financial relationships to disclose
Introduction/Rationale: The hyper-heterogeneity of T cell receptors (TCRs), via V(D)J recombination, enables the immune system to recognize a vast array of pathogens and cancer neoantigens. However, this diversity presents significant computational challenges for precision medicine, including mapping TCRs (particularly their antigen-specific CDR3β regions) to their cognate antigens and modeling the functional impact. Modeling the functional and phenotypical effects of TCR diversity, as well as integrating TCR sequencing data with cellular phenotypes are important active areas of research.
Methods: To address these challenges, we developed TCRClass, a comprehensive framework for immune repertoire analysis and TCR-antigen specificity prediction. In-vivo neoepitope-based immunization assays, using highly specific barcoded multimers, are used to validate the proposed modularity optimization repertoire diversity metric and to train our model. Furthermore, antigen and epitope-specific TCR sequencing of putative cancer neoepitopes presented on MHC-I in mice is used to extract and engineer CDR3 sequence-based features and train the model for TCR specificity prediction.
Results: We developed TCRClass, a comprehensive computational method for analyzing T cell repertoire and predicting TCR-antigen specificity. TCRClass introduces two key innovations: 1) A graph-theoretical algorithm that employs modularity optimization with a split-penalty to identify local, antigen-specific TCR clusters. 2) A novel transfer learning framework that uses pre-trained Protein Language Models (pLMs) to generate powerful embeddings from CDR3 sequences, enabling accurate prediction of TCR specificity for cancer neoantigens. We benchmark state-of-the-art pLMs for this task and demonstrate that TCRClass facilitates integration with scRNA-seq for a unified functional analysis.
Conclusion: Our framework provides a versatile and powerful toolkit for deciphering TCR-antigen interactions, with direct applications in cancer immunology and therapeutic development.
