GTE: a graph learning framework for prediction of T-cell receptors and epitopes binding specificity

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning

Background:

• T-cell receptor (TCR) and peptide (epitope) interactions with major histocompatibility complex (MHC) molecules are central to adaptive immunity.
• Predicting these interactions is vital for understanding immune responses, disease pathogenesis, and developing immunotherapies.

Purpose of the Study:

• To develop a novel computational model, GTE, that leverages the topological structure of TCR-epitope interaction networks for improved prediction accuracy.
• To address limitations of existing sequence-based methods by incorporating network topology.

Main Methods:

• Developed GTE, a heterogeneous Graph neural network model employing inductive learning to capture TCR-epitope interaction topology.
• Implemented a dynamic edge update strategy for effective negative sample construction (non-binding pairs).
• Adapted Deep AUC Maximization for graph-based data imbalance mitigation.

Main Results:

• GTE demonstrated superior performance in predicting TCR-epitope interactions across four public datasets compared to existing methods.
• The model's effectiveness highlights the importance of topological structures in molecular interaction networks.
• The study validates the benefits of analyzing complex molecular network topology.

Conclusions:

• The GTE model offers a powerful new approach for predicting TCR-epitope interactions by effectively utilizing network topology.
• This work underscores the significance of graph-based learning and topological feature extraction in immunoinformatics.
• The findings pave the way for enhanced understanding and manipulation of immune responses.