Yale Researchers Use Machine Learning Tool to Improve Personalized Immunotherapy Design

Yale researchers have developed a machine learning model called ImmunoStruct that could significantly improve the design of personalized cancer vaccines — and the work was supported in part by the Colton Center for Autoimmunity at Yale. The study was published in Nature Machine Intelligence in February 2026.

The tool addresses a key limitation in existing vaccine design models: most treat peptides — the short protein fragments that the immune system recognizes as foreign — as flat, one-dimensional sequences of amino acids. ImmunoStruct instead incorporates three-dimensional structural and biochemical properties of peptides, giving the model a richer, more complete picture of how the immune system is likely to respond.

Epitope-based vaccines, which contain specific peptides designed to trigger targeted immune responses, are an emerging area of cancer immunotherapy showing promise for melanomas, breast cancers, and glioblastomas. By more accurately predicting which peptides will provoke the strongest immune response, ImmunoStruct could help researchers design vaccines tailored to a patient’s unique tumor profile — a potentially less toxic alternative to broad-spectrum therapies like chemotherapy.

The research was led by Professor Smita Krishnaswamy and Professor Akiko Iwasaki of Yale School of Medicine, with co-first authors Kevin B. Givechian and Chen Liu. The team has already licensed ImmunoStruct to Latent-Alpha, a Yale spinout company, and made the model available open source via GitHub.

The study is also notable as a direct output of the 2025 Colton Center at Yale grant cycle, in which ImmunoStruct was among the funded projects.