Deep learning-based prediction of the T cell receptor-antigen binding specificity

Area of Science:

• Immunology
• Computational Biology
• Genomics

Background:

• T cell recognition of tumor cells relies on neoantigens, but predicting which neoantigens elicit responses and their specific T cell receptor (TCR) interactions remains challenging.
• Understanding these interactions is crucial for developing effective cancer immunotherapies.

Purpose of the Study:

• To develop a predictive model (pMTnet) for T cell receptor (TCR) binding specificities to neoantigens and other T cell antigens presented by human leukocyte antigen class I molecules (pMHCs).
• To apply this model to human tumor genomics data to uncover insights into tumor immunogenicity and its correlation with patient prognosis and immunotherapy response.

Main Methods:

• A transfer learning-based deep learning model, pMTnet, was constructed to predict TCR-pMHC binding.
• The model was trained and validated using TCR sequences (CDR3β), antigen sequences, and class I MHC alleles.
• pMTnet was applied to analyze human tumor genomics data from melanoma, lung, and kidney cancers.

Main Results:

• pMTnet demonstrated superior performance in predicting TCR-pMHC binding compared to existing methods.
• Neoantigens were generally found to be more immunogenic than self-antigens across tumor types.
• A specific self-antigen, HERV-E, reactivated in kidney cancer, exhibited higher immunogenicity than neoantigens.
• Stronger T cell expansion and higher affinity for truncal versus subclonal neoantigens correlated with better prognosis and immunotherapy response in melanoma and lung cancer, but not kidney cancer.

Conclusions:

• pMTnet accurately predicts TCR-neoantigen/antigen pairs using minimal sequence and allele information.
• The study provides novel insights into TCR-pMHC interactions within the tumor microenvironment.
• Predictive modeling of TCR-pMHC interactions can serve as a valuable tool for understanding tumor immunogenicity and guiding cancer immunotherapy strategies.