DeepAIR: A deep learning framework for effective integration of sequence and 3D structure to enable adaptive immune...

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Adaptive immune receptors (AIRs), including T cell receptors (TCRs) and B cell receptors (BCRs), bind to antigens, a fundamental process in adaptive immunity.
• Current prediction methods for AIR-antigen binding primarily use sequence data, neglecting crucial structural features that influence binding affinity.

Purpose of the Study:

• To develop a deep learning framework, DeepAIR, that integrates both sequence and structure features for accurate prediction of AIR-antigen binding.
• To improve the prediction accuracy of binding affinity and reactivity for TCRs and BCRs.

Main Methods:

• Developed DeepAIR, a deep learning framework integrating sequence and structure features of AIRs.
• Applied DeepAIR to predict binding affinity of TCRs and binding reactivity of TCRs and BCRs.
• Utilized TCR and BCR repertoire data for disease identification.

Main Results:

• DeepAIR achieved a Pearson's correlation of 0.813 for TCR binding affinity prediction.
• Median AUC values of 0.904 for TCR binding reactivity and 0.942 for BCR binding reactivity were obtained.
• DeepAIR successfully identified all patients with nasopharyngeal carcinoma and inflammatory bowel disease in test data.

Conclusions:

• DeepAIR significantly improves the prediction of AIR-antigen binding by incorporating structural information.
• The framework facilitates a deeper understanding of adaptive immunity.
• DeepAIR demonstrates potential for disease diagnostics by analyzing immune receptor repertoires.