Deciphering small sequence differences in T cell receptor-antigen pairing

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• T cells play crucial roles in immunity and disease via T cell receptor (TCR)-antigen interactions.
• Existing prediction tools often struggle to discern the impact of minor sequence variations in TCRs or antigens on binding affinity.

Purpose of the Study:

• To develop a deep learning model, pMTnet-omni, capable of predicting TCR-pMHC binding and differentiating binding affinities based on sequence similarity.
• To utilize the model's interpretability to uncover biological rules governing TCR-antigen recognition.
• To enable the prediction of variant TCRs with modulated binding strengths for translational applications.

Main Methods:

• Development of a deep learning model (pMTnet-omni) for predicting TCR-pMHC binding.
• Leveraging the model to analyze sequence variations and their effect on binding affinity.
• Integration with a Lab-in-the-Loop (LiL) mechanism for predicting and designing variant TCRs.
• Validation of the model's ability to predict binding for TCRs and pMHCs with sequence similarities.

Main Results:

• pMTnet-omni accurately predicts TCR-pMHC binding and distinguishes between stronger and weaker binding TCRs with similar sequences.
• The model provides insights into the biological determinants of TCR-antigen pairing.
• Accurate prediction of variant TCRs with desired binding characteristics was achieved.
• The model demonstrated efficacy in predicting TCR binding to similar pMHC molecules.

Conclusions:

• pMTnet-omni offers a powerful and flexible toolkit for analyzing TCR-antigen interactions.
• The model facilitates research in immunology and enables translational applications in areas like immunotherapy.
• This approach advances the understanding and engineering of TCR-based therapeutics.