GGNpTCR: A Generative Graph Structure Neural Network for Predicting Immunogenic Peptides for T-cell Immune Response

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• T-cell receptor (TCR) interactions with antigens are vital for adaptive immunity, vaccine design, and immunotherapy.
• Current prediction methods often fail with novel antigens not present in training data and neglect antigen spatial structure.
• Accurate TCR-antigen interaction prediction is essential for developing targeted immunotherapies and vaccines.

Purpose of the Study:

• To develop a novel deep learning framework, GGNpTCR, for predicting T-cell receptor (TCR) and peptide interactions.
• To enhance prediction accuracy for new or unseen antigens and exogenous peptides.
• To incorporate 3D structural information for improved interpretability and precise localization of interactions.

Main Methods:

• Developed GGNpTCR, a deep learning framework utilizing generative graph structures for TCR-peptide interaction prediction.
• Employed sequence information as primary input for interaction prediction.
• Integrated supervised mechanisms to forecast interaction locations within 3D configurations.

Main Results:

• GGNpTCR demonstrated excellent prediction performance on new antigens absent from training datasets, outperforming existing methods.
• Significant improvements were observed when applying the model to a large COVID-19 dataset with novel antigens.
• Incorporation of 3D structural information enhanced the model's interpretability by precisely forecasting interaction locations.

Conclusions:

• GGNpTCR represents a significant advancement in predicting TCR-antigen interactions, offering improved performance and universality.
• The model's ability to generalize to unseen antigens is crucial for developing next-generation vaccines and immunotherapies.
• Enhanced interpretability through 3D structural analysis provides deeper insights into TCR-antigen binding mechanisms.