Abstract T cells are essential contributors to anti-tumor immunity, but identifying the specific T-cell receptor (TCR) features that differentiate tumor-reactive responses from background immune activity remain difficult to resolve. Tumors reshape TCR repertoires through neoantigen exposure, chronic stimulation, and immune escape, creating patterns that are biologically relevant not easily quantified with existing computational methods. Current approaches either rely on global repertoire summaries or on deep learning models with limited interpretability, which restricts their translational and clinical applications.To address this gap, we developed HELP-TCR, an explainable machine learning framework that applies concepts from natural language processing to the functional analysis of tumor-associated TCR repertoires. HELP-TCR represents TCR repertoires by the position-specific distributions of features (single amino acids and/or amino acid pairs), transforming sequences into tensor structures. To increase reproducibility, we adapt the principles of ensemble learning and proposed a series of preprocessing steps, among which a consensus grouping method is applied to merge the features with highly similar position-wise distributions opening the venue for explainable dimension reduction. A modified (to the dimensionality of the task) deep learning architecture enables accurate classification, while post-hoc analysis based on saliency map highlights the most informative features contributing to model predictions.Using a dataset of TCR sequences from non–small cell lung cancer, HELP-TCR demonstrated stable and high predictive performance (AUC ∼0.96), outperforming DeepTCR (AUC 0.76) and TCR-BERT embeddings, which showed limited class separability. Post-hoc saliency analysis identified positional amino acid motifs and residue-pair interactions that contributed to tumor–normal discrimination, offering mechanistic insights potentially reflecting tumor reactivity, immune pressure, or clonal remodeling within the tumor microenvironment.HELP-TCR offers an interpretable and reproducible framework for analyzing tumor-associated TCR repertoires in settings where sequence-level information is relevant. Its representations align with emerging clinical applications of TCR profiling, such as assessing of response to immunotherapy, identification of neoantigen-reactive or tumor-enriched clonotypes and studying of immune dynamics within the tumor tissue. HELP-TCR provides a practical foundation for generating testable hypotheses and advancing translational efforts aimed at improving patient stratification and informing T-cell–based therapeutic strategies. Citation Format: Yulyana Kalesnik, Dawid Krawczyk, Maciej Pietrzak, Michal Seweryn. HELP-TCR: A harmonized and explainable language processing framework for functional analysis of t-cell receptor repertoires abstract. In: Proceedings of the American Association for Cancer Research Annual Meeting 2026; Part 1 (Regular Abstracts); 2026 Apr 17-22; San Diego, CA. Philadelphia (PA): AACR; Cancer Res 2026;86(7 Suppl):Abstract nr 2695.
Kalesnik et al. (Fri,) studied this question.