Improved Deep Learning Prediction of TCR-HLA Association

Abstract

Understanding the relationship between T cell receptors (TCRs) and human leukocyte antigens (HLAs) is essential for elucidating immune response specificity, uncovering mechanisms of autoimmunity, and advancing targeted immunotherapies. We have previously developed a deep learning method, DePTH (Deep Learning Prediction of TCR-HLA associations), to predict the association between a TCR and an HLA based on their amino acid sequences. In this work, we evaluated the performance of DePTH in two additional datasets, and investigated the influence of two potential confounding factors: TCR generation probability and the sequence length of CDR3 (Complementarity-Determining Region 3), which is a key region in the antigen-binding site of TCRs. Building on these insights, we combined training data from two datasets to train a new version of DePTH: DePTH 2.0.