Researchers at Radboudumc have developed SwiftMHC, a new AI tool that can predict how the immune system recognizes harmful substances much faster than before. The study, conducted with the Netherlands eScience Center, has been published in Cell Reports Methods on March 31.
Why this matters
Our immune system relies on molecules that show small fragments of viruses or cancer cells to immune cells. These fragments, called peptides, need to bind to so-called MHC molecules for the immune system to detect them. Understanding this process is essential for understanding auto-immunity, developing vaccines and improving cancer treatments.
Until now, predicting how these molecules fit together in 3D has been slow and required a lot of computing power. This has limited large-scale use.
A faster way to model the immune system
SwiftMHC uses artificial intelligence to predict how peptides and MHC molecules fit together in three dimensions, and their binding affinity, in milliseconds when running in batch mode. What used to take much longer can now be done in milliseconds, without losing accuracy. As Li Xue, who led the study, explains: “We wanted to build a tool that is both fast and precise, so researchers can explore immune interactions at a scale and speed that wasn’t possible before.”
Why it stands out
SwiftMHC combines speed and detail. It is as fast as simpler prediction tools, but also provides detailed 3D models that researchers can use to better understand immune responses. At the same time, it requires far less computing power, making it a more sustainable approach.
The researchers will be working with partners to test the predictions in the wet-lab. The goal is to improve the safety and effectiveness of new immune-based therapies.
This work marks an important step toward faster and more efficient tools to study and improve the immune system.
Video | Demonstration of the learning process of SwiftMHC. This animation shows how SwiftMHC figures out how a small piece of a virus or cell (peptide) fits into an immune protein. It starts as a rough shape and gradually folds into the right position, step by step, until a realistic structure is formed.
About the publication
Baakman C., Crocioni G., Geng C., Rademaker D. T., Frühbuß D., Aarts Y.J.M., Xue L. C. A high-speed attention network for MHC-bound peptide identification and 3D modeling. Cell Reports Methods, 2026; 0. DOI: 10.1016/j.crmeth.2026.101364
