RACER-m leverages structural features for sparse T cell specificity prediction.

Area of Science:

• Immunology
• Computational Biology
• Structural Biology

Background:

• Predicting T cell receptor (TCR) specificity against antigens is complex due to extensive sequence diversity and limited training data.
• Existing models struggle to accurately predict MHC-allele-restricted TCR-peptide interactions, despite incorporating structural information.

Purpose of the Study:

• To develop a coarse-grained structural model (RACER-m) for predicting TCR-antigen recognition.
• To leverage biophysical information from TCR-antigen crystal structures to improve prediction accuracy and reduce data requirements.

Main Methods:

• Developed RACER-m, a coarse-grained structural model incorporating biophysical insights from TCR-antigen crystal structures.
• Utilized publicly available structural data for model training and validation.

Main Results:

• Explicit inclusion of structural data significantly reduced the number of training examples needed.
• The model demonstrated reliable predictions of TCR recognition specificity and sensitivity across various contexts.
• RACER-m successfully identified point-mutant peptides affecting binding affinity, outperforming sequence-based methods.

Conclusions:

• RACER-m offers a robust approach for predicting TCR-antigen interactions, overcoming limitations of sequence-based methods.
• The model's reliance on structural information makes it broadly applicable to diverse TCR-peptide pairs.
• This approach can aid in understanding TCR specificity and designing immunotherapies.