Modeling TCR-pMHC Binding with Dual Encoders and Cross-Attention Fusion.

Area of Science:

• Computational immunology
• Bioinformatics
• Immunoinformatics

Background:

• Accurate modeling of T-cell receptor (TCR) and peptide-MHC (pMHC) interactions is crucial for developing immunotherapies and personalized vaccines.
• The high diversity of TCR repertoires and limited experimental data pose challenges for predicting TCR-pMHC binding to new epitopes.

Purpose of the Study:

• To develop TIDE, a cross-attention-driven dual-encoder framework for enhanced TCR-pMHC binding prediction.
• To leverage large protein and molecular language models for learning discriminative TCR and peptide representations.

Main Methods:

• TIDE utilizes Evolutionary Scale Modeling (ESM) for TCR sequence encoding and MolFormer for peptide SMILES string processing.
• A multi-layer cross-attention mechanism integrates TCR and peptide embeddings to identify interaction patterns without structural alignment.
• The framework was evaluated on the TCHard benchmark in zero-shot and few-shot learning scenarios.

Main Results:

• TIDE demonstrated superior predictive accuracy and robustness compared to existing methods like ChemBERTa, TITAN, and NetTCR.
• The cross-attention fusion of pretrained language models effectively captures TCR-pMHC binding determinants.
• The model shows strong generalization capabilities for unseen epitopes.

Conclusions:

• Combining pretrained language models with cross-attention fusion provides a powerful approach for TCR-pMHC binding prediction.
• TIDE offers a reliable computational tool for advancing immunotherapy and vaccine design.
• This work paves the way for more robust applications in computational immunology.