A functionally validated TCR-pMHC database for TCR specificity model development

Area of Science:

• Immunology
• Computational Biology
• Bioinformatics

Background:

• Predicting T-cell receptor (TCR) specificity is crucial in immunology.
• Current TCR-pMHC prediction models rely on historical data of variable quality.
• Large language models and computational structure prediction offer potential solutions.

Purpose of the Study:

• To develop and employ a high-throughput synthetic platform for validating TCR-pMHC interactions.
• To assess the functional accuracy of entries in the VDJdb database.
• To generate a high-quality, validated TCR database (TCRvdb) for community use.

Main Methods:

• Established a high-throughput synthetic platform for TCR assembly and functional evaluation.
• Assessed a large subset of VDJdb-deposited TCR-pMHC entries using standardized functional readouts.
• Utilized TCRbridge to analyze AlphaFold3 confidence metrics for TCR-pMHC predictions.

Main Results:

• Confirmed functional TCR reactivity for only 50% of the evaluated VDJdb entries.
• Demonstrated AlphaFold3 confidence metrics can distinguish functionally validating from non-validating TCRs, despite not being trained for this task.
• Generated the validated TCRvdb database.

Conclusions:

• A significant portion of existing TCR-pMHC data lacks functional validation.
• The validated TCRvdb database and AlphaFold3 confidence metrics show promise for improving TCR specificity prediction.
• This resource will aid in training and evaluating more accurate predictive TCR specificity models.