Explainability from Training with Applications to TCR-Epitope Prediction
Abstract
Deep learning models have achieved strong performance in artificial intelligence for science, yet their black-box nature limits our understanding of how they learn scientific tasks. Existing methods for interpretability provide limited insight into how models organize evidence and evolve during learning. We introduce explainability from training (EFT), a model-agnostic paradigm that traces model interpretation during training to explain why models rely on specific features and how they organize these features as predictive evidence. We apply EFT to four state-of-the-art T cell receptor (TCR)-epitope prediction models, TCR-SRIM, TULIP, MixTCRpred, and NetTCR-2.2, spanning post-hoc and interpret-by-design approaches as well as transformers and CNNs. To investigate how structural information affects model explanations, we introduce a benchmark, TCR-XAI2, containing 388 unique experimentally resolved TCR-epitope structures, complemented by structures predicted using AlphaFold3, Boltz-2, TCRModel2, tFold-TCR, and OpenFold3. Using EFT with TCR-XAI2, we demonstrate that (1) CNN and transformer models exhibit distinct learning trajectories; (2) TCR α and β evidence can conflict during learning, limiting the benefits of jointly modeling both chains, while MHC information mitigates this; and (3) real versus predicted structural data for TCR-epitope prediction exhibits distinct TCR and peptide feature preferences as well as differing trajectories of model certainty.
Related
- Enhancing Linear B-Cell Epitope Prediction via Evolutionary Scale Modeling (ESM-2) and Hybrid Feature Integration
- Improving RNA secondary structure prediction through expanded training data
- Metabolomics in the Prediction and Early Diagnosis of Type 2 Diabetes Mellitus: Advances and Applications.
- Deep learning-driven classification of 46 medicinal macrofungi species: comparative analysis, fusion strategies, and explainability insights.
- Conservation guided reverse vaccinology for the design of a broadly immunogenic multi-epitope vaccine against monkeypox virus
BibTeX
@misc{li2026explainability,
title = {{Explainability from Training with Applications to TCR-Epitope Prediction}},
author = {Jiarui Li and Zixiang Yin and Samuel Landry and Zhengming Ding and Ramgopal Mettu},
howpublished = {arXiv},
year = {2026},
month = sep,
doi = {10.48550/arxiv.2609.36354},
eprint = {2609.36354},
archivePrefix = {arXiv},
url = {https://arxiv.org/abs/2609.36354},
note = {CC BY 4.0},
}
RIS
TY - UNPB
TI - Explainability from Training with Applications to TCR-Epitope Prediction
AU - Jiarui Li
AU - Zixiang Yin
AU - Samuel Landry
AU - Zhengming Ding
AU - Ramgopal Mettu
PY - 2026
DA - 2026/09/28
PB - arXiv
DO - 10.48550/arxiv.2609.36354
UR - https://arxiv.org/abs/2609.36354
AB - Deep learning models have achieved strong performance in artificial intelligence for science, yet their black-box nature limits our understanding of how they learn scientific tasks. Existing methods for interpretability provide limited insight into how models organize evidence and evolve during learning. We introduce explainability from training (EFT), a model-agnostic paradigm that traces model interpretation during training to explain why models rely on specific features and how they organize these features as predictive evidence. We apply EFT to four state-of-the-art T cell receptor (TCR)-epitope prediction models, TCR-SRIM, TULIP, MixTCRpred, and NetTCR-2.2, spanning post-hoc and interpret-by-design approaches as well as transformers and CNNs. To investigate how structural information affects model explanations, we introduce a benchmark, TCR-XAI2, containing 388 unique experimentally resolved TCR-epitope structures, complemented by structures predicted using AlphaFold3, Boltz-2, TCRModel2, tFold-TCR, and OpenFold3. Using EFT with TCR-XAI2, we demonstrate that (1) CNN and transformer models exhibit distinct learning trajectories; (2) TCR α and β evidence can conflict during learning, limiting the benefits of jointly modeling both chains, while MHC information mitigates this; and (3) real versus predicted structural data for TCR-epitope prediction exhibits distinct TCR and peptide feature preferences as well as differing trajectories of model certainty.
ER -