Researchers at the Translational Genomics Research Institute (TGen), part of City of Hope , have developed a new approach using the artificial intelligence platform AlphaFold3 to predict whether a T cell receptor will recognize a specific antigen based solely on its genetic sequence. The advancement could help scientists more rapidly decode immune responses to infectious diseases, tumors and other health threats.
Published in Science Advances , the study demonstrates how AlphaFold3 can accurately model a complex three-part molecular interaction that lies at the center of the body's adaptive immune response. By revealing how these molecules fit together, the approach may help researchers better predict which immune cells can recognize and respond to specific pathogens or cancer targets.
Understanding the Immune System's Recognition Process
For your immune system to fight off a threat, three things must work together: a display molecule called MHC, a small piece of the invader called a peptide, and a receptor on a T cell called a TCR.
It starts when an MHC molecule grabs a fragment of a virus, bacterium, or tumor cell and holds it out for inspection. A T cell then checks that fragment with its receptor. If the receptor fits the MHC and its fragment just right, the T cell knows it has found a threat and launches an attack aimed at it.
Scientists call this triad, a set of three molecules that only work as a team, the MHC:peptide:TCR complex. It is one of the main ways the immune system recognizes what it needs to fight.
Using AI to Solve a Long-Standing Challenge
The unique structure of each T cell receptor enables the immune system to distinguish among billions of potential antigens. While scientists have experimentally determined the structures of some TCRs and the antigens they recognize, many others are known only through their genetic sequences.
Until now, predicting whether a given T cell receptor would recognize an arbitrary antigen has remained a significant challenge.
"Before this, if you didn't know anything about the target antigen and you wanted to know if a TCR sequence was specific to it or not, the best you could come up with was a roll-of-the-dice performance, or random guess," said John Altin, Ph.D. , an associate professor in TGen's Immunology and Microbiome Division and a senior author of the study. "Now we can do clearly better than chance. It's remarkable that a general structural model like AlphaFold 3 works on such a specialized task, and we expect it to improve from here."
Harnessing the Power of AlphaFold3
AlphaFold transformed structural biology by enabling researchers to predict three-dimensional molecular structures from genetic sequences. Introduced in 2024, AlphaFold3 was designed specifically to predict interactions among multiple molecules, making it particularly well suited for modeling MHC:peptide:TCR complexes.
To evaluate the platform's performance, the TGen team first tested AlphaFold3 using triads whose structures had already been determined experimentally. After confirming its accuracy, the researchers analyzed a larger group of immune complexes to identify structural features associated with T cell recognition and non-recognition of specific antigens.
The investigators then challenged the model using high-quality genetic sequences from previously uncharacterized immune complexes, including eight MHC class II-antigen pairs derived from the bacterium that causes tuberculosis.
Tuberculosis presented an especially rigorous test because the pathogen contains numerous antigens, while the MHC molecules responsible for presenting those antigens vary significantly among individuals and populations worldwide. It is also a disease where new vaccines and diagnostics are sorely needed.
"These antigens fall within large unexplored or underexplored regions of this space, something we are very interested in," said Altin.
Potential Applications in Infectious Disease and Cancer Research
The researchers found that AlphaFold3 was able to model these immune complexes with sufficient accuracy to predict whether a T cell receptor would recognize antigens known only from their genetic sequences.
If further validated, the approach could help accelerate research into infectious diseases, cancer immunotherapy and vaccine development by enabling scientists to identify promising immune targets more efficiently.
By helping researchers better understand how immune cells recognize threats, the technology may provide a powerful new tool for studying the immune responses that protect the body from disease.
The researchers have been working simultaneously on a second triad challenge, said Kamel Lahouel, Ph.D. , an assistant professor in TGen's Early Detection and Prevention Division and co-author of the study. "We have also been very interested in the particular step that comes before the TCR gets involved, which is how to predict the binding affinity between the peptide and an MHC class II molecule."
In a joint effort across TGen research groups involving teams led by Lahouel, Altin, and Cristian Tomasetti, Ph.D. , the researchers are building on AlphaFold3's structural modeling with downstream graph-based neural networks, trained using data on hundreds of thousands of peptide interactions from Altin's PepSeq assays, to develop a predictor of MHC class II–peptide binding affinity.
