Track genomics, neuroscience and molecular biology.
Search papers, labs, and topics across Lattice.
Explainability from Training (EFT) is introduced as a model-agnostic paradigm to trace interpretability during deep learning training, revealing how models learn and organize features for scientific tasks. Applied to TCR-epitope prediction models using a novel structural benchmark (TCR-XAI2), EFT uncovers distinct learning trajectories for CNNs and transformers, highlights conflicts between TCR $\alpha$/$\beta$ evidence that MHC information can resolve, and differentiates model behavior based on real versus predicted structural data. These findings provide granular insights into model learning dynamics and feature reliance in complex biological prediction tasks.
Real versus predicted TCR-epitope structures lead to fundamentally different model certainty and feature preferences, suggesting structural data quality is a critical bottleneck for AI in science.
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 $\alpha$ and $\beta$ 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.