TCR-EML: Explainable Model Layers for TCR-pMHC Prediction.

Area of Science:

• Immunology
• Computational Biology
• Machine Learning

Background:

• T cell receptor (TCR) recognition of peptide-MHC (pMHC) complexes is crucial for adaptive immunity, impacting vaccines, cancer immunotherapy, and autoimmune diseases.
• Current machine learning models for TCR-pMHC binding prediction are often "black-box," lacking mechanistic explanations.
• Existing post-hoc explanation methods offer limited insight into biochemical binding mechanisms.

Purpose of the Study:

• To introduce explainable model layers (TCR-EML) for TCR-pMHC binding prediction.
• To integrate explainable components into protein-language model backbones for interpretable TCR-pMHC modeling.
• To enhance the transparency and interpretability of TCR-pMHC binding predictions.

Main Methods:

• Developed TCR-EML, an "explain-by-design" approach incorporating prototype layers for amino acid residue contacts.
• Utilized known TCR-pMHC binding mechanisms to inform the model architecture.
• Incorporated TCR-EML into protein-language model backbones for TCR-pMHC prediction.

Main Results:

• Achieved competitive predictive accuracy and generalization on large-scale datasets.
• Demonstrated high-quality explanations for predicted TCR-pMHC binding.
• Showcased improved explainability compared to existing methods on the TCR-XAI benchmark.

Conclusions:

• TCR-EML offers a novel "explain-by-design" solution for interpretable TCR-pMHC binding prediction.
• The method balances predictive performance with mechanistic interpretability.
• This approach has significant potential for advancing vaccine design, cancer immunotherapy, and understanding autoimmune diseases.