GATTCR: A Graph Attention Network With Multi-Feature Fusion for Peripheral Blood TCR Repertoire Classification.

Area of Science:

• Immunoinformatics
• Computational Biology
• Machine Learning

Background:

• T cell receptor (TCR) repertoire profiling from peripheral blood offers noninvasive disease diagnostics.
• High diversity and sparsity of TCR sequences challenge accurate immune state classification.

Purpose of the Study:

• To introduce GATTCR, a novel framework integrating Graph Attention Networks (GATs) and multi-feature fusion for TCR repertoire analysis.
• To improve the accuracy and generalizability of immune state classification using TCR data.

Main Methods:

• Representing TCR sequences as graph nodes within a Graph Attention Network (GAT) framework.
• Incorporating biological priors: sequence embeddings, structural similarity, V gene usage, and clonal frequency.
• Multi-feature fusion for context-aware, structure-informed representation learning.

Main Results:

• GATTCR demonstrated consistent performance improvements across diverse cancer and infectious disease datasets.
• Achieved significant Area Under the Receiver Operating Characteristic Curve (AUROC) gains of up to +47.9% in few-shot learning scenarios.
• Ablation studies confirmed the importance of graph-based modeling and immunological features.

Conclusions:

• GATTCR provides a scalable and effective approach for TCR repertoire analysis.
• The framework enhances noninvasive immune monitoring capabilities for precision medicine.
• GATTCR shows strong potential for generalizing from limited labeled data in disease diagnostics.