TPBTE: A model based on convolutional Transformer for predicting the binding of TCR to epitope

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning

Background:

• T cell receptors (TCRs) are crucial for adaptive immunity, recognizing specific antigens via their unique sequences.
• Current TCR-epitope binding prediction methods often overlook long-range amino acid interactions and inter-sequence relationships, leading to dataset-dependent variability.
• Accurate prediction of TCR-epitope interactions is essential for understanding immune responses and developing immunotherapies.

Purpose of the Study:

• To introduce TPBTE, a convolutional Transformer-based model for predicting T cell receptor (TCR) to epitope binding.
• To improve the accuracy of TCR-epitope binding predictions by incorporating local sequence features and inter-sequence interactions.
• To provide a computational tool that can aid in the initial stages of epitope screening.

Main Methods:

• TPBTE utilizes epitope sequences and TCRβ complementary determining region 3 (CDR3) sequences as input.
• A convolutional attention mechanism is employed to learn amino acid representations by capturing local sequence features.
• Cross-attention mechanisms are used to model the interaction information between TCR and epitope sequences.

Main Results:

• TPBTE demonstrated superior performance compared to baseline models, evidenced by a higher average area under the curve on TCR-epitope binding data.
• The model effectively learns amino acid representations and captures inter-sequence interactions.
• TPBTE provides binding probabilities, facilitating efficient epitope screening.

Conclusions:

• TPBTE offers an advanced approach for predicting TCR-epitope binding, outperforming existing methods.
• The model's ability to consider complex sequence relationships enhances prediction accuracy.
• TPBTE can significantly accelerate epitope screening processes, reducing search time and scope.