AI tool AlphaFold3 can now predict how T cells recognize threats — and it could transform immunotherapy

Scientists at the Translational Genomics Research Institute (TGen), part of City of Hope, have found a powerful new use for AlphaFold3 — the AI platform already renowned for predicting protein structures. Published in Science Advances, the study shows that AlphaFold3 can model the three-molecule complex (MHC + peptide + T cell receptor) that sits at the core of how the immune system identifies threats, and do so accurately enough to predict T cell recognition from genetic sequences alone.

This is a significant advance because, until now, predicting whether a T cell receptor would recognize a given antigen was essentially a coin flip. The team validated the approach using experimentally confirmed immune complexes, then stress-tested it on tuberculosis antigens — a notoriously complex and underexplored target where MHC molecules vary widely across populations.

Key Takeaways:

• AlphaFold3 can predict T cell receptor–antigen recognition better than chance, using only genetic sequence data
• The model was tested on TB antigens, where MHC molecules vary widely across populations — a tough real-world challenge
• Researchers are now building on this with graph-based neural networks to predict MHC class II–peptide binding affinity

Why it matters: Identifying which immune cells can recognize specific pathogens or tumors is foundational to designing better vaccines and cancer immunotherapies. This AI-driven approach could dramatically accelerate that discovery process.