ATM-TCR: TCR-Epitope Binding Affinity Prediction Using a Multi-Head Self-Attention Model.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• T cell receptor (TCR)-epitope binding is central to T cell regulation and immunotherapy development.
• Current computational methods for predicting TCR-epitope binding struggle with limited epitope data, leading to poor generalization.
• Accurate prediction is essential for understanding T cell interactions and advancing therapeutic strategies.

Purpose of the Study:

• To develop a novel computational model, ATM-TCR, for predicting TCR-epitope binding.
• To improve the generalization performance of TCR-epitope binding prediction models, especially for unseen epitopes.
• To leverage attention mechanisms for enhanced biological contextual understanding and improved predictive power.

Main Methods:

• Utilized a multi-head self-attention mechanism within the ATM-TCR model.
• Captured biological contextual information from TCR and epitope sequence data.
• Applied attention maps for improved out-of-sample performance analysis.

Main Results:

• ATM-TCR demonstrated improved generalization performance compared to existing methods.
• The model effectively captured biological contextual information relevant to TCR-epitope binding.
• Demonstrated successful application on SARS-CoV-2 data, highlighting potential for real-world impact.

Conclusions:

• The ATM-TCR model offers a promising approach for accurate TCR-epitope binding prediction.
• The multi-head self-attention mechanism enhances model generalization, addressing limitations of current methods.
• Novel application of attention maps provides insights and improves prediction for novel epitopes, particularly relevant for infectious diseases like COVID-19.