Mitigating negative data bias to enhance TCR-epitope binding and residue interaction prediction.

Negative data augmentation with region-directed mutations improves TCR-epitope binding and interaction prediction, enhancing model discrimination and robustness.

• Why it matters: Accurate prediction of TCR-epitope interactions is essential for immunotherapy and vaccine design, but current methods struggle with negative sample quality and interpretability.
• What they did: Researchers developed a negative dataset construction strategy using targeted mutations in the CDR3β region and built TranTCR, comprising two models: TranTCR-bind for binding probability and TranTCR-map for residue-level interaction analysis.
• The result: TranTCR models outperform existing approaches in prediction accuracy and generalization, with TranTCR-map revealing detailed amino acid interactions and cross-reactivity, advancing immune recognition understanding.