General prediction of T cell receptor antigen specificity from sequence using AlphaFold 3

Area of Science:

• Immunology
• Structural Biology
• Bioinformatics

Background:

• The Major Histocompatibility Complex (MHC):peptide:T Cell Receptor (TCR) complex is crucial for adaptive immunity.
• Decoding TCR specificity against diverse antigens remains a significant challenge in immunology.
• Current *in silico* methods struggle to predict TCR interactions with novel or 'unseen' epitopes.

Purpose of the Study:

• To evaluate the accuracy of AlphaFold 3 (AF3) in predicting the structures of MHC:peptide:TCR complexes.
• To develop a predictive model for TCR specificity using AF3-derived structural features.
• To assess the model's ability to generalize predictions to unseen epitopes and TCRs.

Main Methods:

• Application of AlphaFold 3 to predict structures of >9,000 TCRs binding to >1,000 epitopes across >70 MHC alleles.
• Identification of structural features distinguishing cognate (binding) MHC:peptide:TCR triads from non-cognate ones.
• Validation of a predictive model using unseen triads, assessing performance via Area Under the Curve (AUC).

Main Results:

• AlphaFold 3 demonstrated unprecedented accuracy in predicting MHC:peptide:TCR complex structures.
• The developed model achieved high median AUCs (0.81-0.92) on validation datasets.
• The model successfully predicted TCR specificity for epitopes not encountered during training or feature selection.

Conclusions:

• Generalized prediction of TCR specificity from sequence is achievable.
• AlphaFold 3 significantly advances the ability to decode TCR specificity.
• This approach holds potential to accelerate the understanding and decoding of immune responses.