nuTCRacker: Predicting the Recognition of HLA-I-Peptide Complexes by αβTCRs for Unseen Peptides

Area of Science:

• Immunology
• Bioinformatics
• Computational Biology

Background:

• Accurate prediction of T-cell receptor (TCR) and antigenic peptide interactions is vital for understanding T-cell repertoire selection and developing cell-mediated immunotherapies.
• Current prediction methods struggle with peptides not included in their training data, limiting their clinical applicability.

Purpose of the Study:

• To develop and evaluate a novel deep learning method, nuTCRacker, for predicting the recognition of antigenic peptides by CD8+ T-cell receptors (TCRs).
• To assess the performance of nuTCRacker on unseen peptides and identify conditions under which accurate predictions can be made.

Main Methods:

• Development of a deep learning model, nuTCRacker, for predicting TCR-peptide interactions.
• Evaluation using a large dataset from curated public resources and a small, cellula-validated dataset of cancer-associated TCR-peptide pairs.
• Analysis of factors influencing prediction accuracy for unseen peptides.

Main Results:

• nuTCRacker achieved an Area Under the Curve (AUC) > 0.7 for approximately one-third of unseen peptides evaluated on a large dataset.
• The model demonstrated utility in predicting interactions for unseen peptides when the training data included similar HLA class I molecules, peptides, and TCRs.
• Successful evaluation on a small, cellula-validated dataset of cancer-associated TCR-peptide interactions.

Conclusions:

• nuTCRacker represents a significant advancement in predicting TCR-peptide recognition, offering potential for targeted immunotherapies.
• Prediction accuracy for unseen peptides is achievable under specific data conditions, highlighting the importance of diverse and relevant training datasets.
• The findings pave the way for improved T-cell repertoire analysis and the design of novel cancer immunotherapies.