THREADS ties together pathology and molecular biology to answer wide-ranging cancer questions

Faisal Mahmood, PhD

Long Phi Le, MD, PhD

As AI advances, it can help answer more questions about a patient’s cancer and prognosis. But the more specific and clinically interesting questions become, the fewer examples there are to train an AI model. Investigators at Mass General Brigham and collaborating institutions set out to build an AI model that already understands enough about what cancer tissue looks like and what that appearance means biologically that it can help answer questions about cancer pathology even when large reference sets aren’t available.

The new foundation model, known as THREADS, was pretrained on more than 47,000 pathology slides and then tested across 107 oncology tasks. During pretraining, THREADS learned to connect patterns in pathology slides with corresponding molecular profiles, helping the model recognize features that reflect the underlying biology of the tissue.

Researchers compared its performance against leading pathology AI models for a range of questions related to cancer subtypes and grading, tumor microenvironment, molecular alterations and biomarkers, response to treatment, survival and more. THREADS outperformed the other models overall, performing particularly well when limited data were available. The research team included co-lead authors Anurag Vaidya, PhD, Andrew Zhang, and Guillaume Jaume, PhD, with co-senior authors Long Phi Le, MD, PhD, and Faisal Mahmood, PhD, all of the Mass General Brigham Department of Pathology.

Their results establish molecularly guided pretraining as a potentially powerful path toward building versatile pathology AI models that can tackle a wide range of clinically meaningful oncology questions, including questions for which limited data are available.

Published in Nature Cancer on November 20, 2026 | Read the paper: “A Multimodal Foundation Model for Molecular and Morphologic Data in Oncology”

Summary reviewed by: Anurag Vaidya, PhD, and Andrew Zhang, co-lead authors; Long Phi Le, MD, PhD, and Faisal Mahmood, PhD, co-senior authors

Leave a Comment