SageTCR: a structure-based model integrating residue- and atom-level representations for enhanced TCR-pMHC binding...

Area of Science:

• Immunology
• Computational Biology
• Structural Biology

Background:

• T-cell receptors (TCRs) are crucial for adaptive immunity, recognizing peptide-MHC (pMHC) complexes.
• The inherent diversity and cross-reactivity of TCRs present significant challenges for predicting TCR-epitope interactions and developing TCR-based therapies.

Purpose of the Study:

• To introduce SageTCR, a bi-level graph neural network (GNN) framework designed for predicting TCR-pMHC binding possibilities.
• To leverage structural data and pretrained language models for enhanced prediction accuracy.

Main Methods:

• SageTCR employs a GNN architecture to encode structural information at both residue and atomic levels.
• Attention mechanisms integrate bimodal representations, and data augmentation strategies address the scarcity of experimental structures.
• The framework preserves the characteristic diagonal binding mode of TCR-pMHC interactions.

Main Results:

• SageTCR significantly outperforms six other deep learning methods in predicting TCR-pMHC binding.
• The model demonstrates interpretability by identifying key contact residues and their conformational features at the interface.

Conclusions:

• SageTCR provides a powerful and accurate method for predicting TCR-pMHC interactions, advancing TCR-related therapies.
• The framework's interpretability offers valuable insights for TCR engineering and the design of immunotherapies.