TGen Models T Cell Recognition With AlphaFold3

Researchers at the Translational Genomics Research Institute tested AlphaFold3 on immune complexes linked to tuberculosis, a step toward predicting which antigens T cells recognize.

Updated on Nov. 4, 2026 in Allergies

TGen Models T Cell Recognition With AlphaFold3

Researchers at the Translational Genomics Research Institute developed an AlphaFold3-based approach to predict whether a T cell receptor recognizes an antigen known only from its genetic sequence. The study was published in Science Advances.

Why it matters

Predicting T cell recognition of arbitrary antigens has been a significant challenge, and the approach could help inform research into infectious disease, cancer immunotherapy, and vaccine development if further validated. New vaccines and diagnostics for tuberculosis are sorely needed.

The test used eight MHC class II-antigen pairs derived from the bacterium that causes tuberculosis, and a companion binding predictor is being trained on hundreds of thousands of peptide interactions. The reported results do not yet quantify prediction accuracy for these pairs, and the approach requires further validation.

The players

Translational Genomics Research Institute (TGen)

A Phoenix-based research institute focused on genomics and translational medicine, where the study was conducted.

AlphaFold3

A structural prediction model introduced in 2024 by Google DeepMind and Isomorphic Labs that predicts interactions among multiple molecules.

The details

The researchers first tested AlphaFold3, introduced in 2024 to predict interactions among multiple molecules, on immune complexes whose structures had already been determined experimentally. They then applied it to high-quality genetic sequences from previously uncharacterized immune complexes to challenge the model. TGen is building a graph-based neural network to predict MHC class II-peptide binding affinity.

AlphaFold3 was introduced in 2024.
TGen's T cell recognition research was reported on November 3, 2026.

Big Picture

The work extends AlphaFold3, a 2024 model built to predict interactions among multiple molecules, into immune recognition, a problem that has resisted prediction for years. It pairs structural modeling with graph-based neural networks trained on hundreds of thousands of peptide interactions, a combined approach that could shape how vaccine and immunotherapy research is conducted.

The approach is still at the research stage and has not been reported as a treatment or diagnostic available to patients. Any future vaccine or diagnostic built on this kind of prediction would need further validation before reaching clinical use.

The takeaway

Predicting which antigens a T cell will recognize is central to vaccine design, and this study offers a computational route to that prediction that does not depend solely on experimentally solved structures. Its value will depend on validation beyond the initial tuberculosis-derived test set.