Comprehensive epitope mutational scan database enables accurate T cell receptor cross-reactivity prediction.

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Predicting T cell receptor (TCR) activation is difficult due to limited datasets and computational methods insensitive to minor peptide mutations.
• Accurate prediction is crucial for understanding immune responses and developing TCR-based therapies.

Purpose of the Study:

• To create a comprehensive dataset for benchmarking TCR activation prediction.
• To develop a novel, interpretable computational model for predicting TCR-peptide interactions.
• To enable efficient learning of TCR binding profiles for novel TCRs.

Main Methods:

• Curated BATCAVE database: >22,000 TCR-peptide pairs from 25 epitopes and 151 TCRs.
• Developed BATMAN: an interpretable Bayesian model for predicting peptide activation of TCRs.
• Implemented active learning for efficient novel TCR profile learning.

Main Results:

• BATMAN outperforms existing methods in predicting TCR-peptide interactions.
• Identified key biochemical predictors of TCR-peptide binding.
• Demonstrated BATMAN's utility in predicting off-target effects and polyclonal T cell responses.

Conclusions:

• BATMAN provides a robust and interpretable framework for predicting TCR activation.
• The BATCAVE database and BATMAN model advance TCR-peptide interaction research.
• Applications include enhancing TCR-based therapies and understanding immune responses.