Ai2 and Providence Swedish Cancer Institute partner to advance AI-assisted scientific discovery
Allen Institute (AI2)
Ai2’s cancer AI is moving into Providence Swedish’s own research data. It just helped spot a breast-cancer immune signal researchers then checked in the lab.
Based on reporting by Allen Institute (AI2) — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Ai2 has moved its AutoDiscovery platform into a new setting: the Paul G. Allen Research Center at Providence Swedish Cancer Institute will use it across cancer research datasets. That matters because this is no longer just a public-data experiment. It’s now inside an active oncology program, where the bar for anything called “AI” is much higher than a clever demo and a nice slide deck.
The trigger for the expansion was a paper built by a joint team from Providence and Ai2, led by Dr. Kelly Paulson, Dr. Sasha Stanton, and Ai2 senior research scientist Bodhisattwa Majumder. The group used AutoDiscovery on The Cancer Genome Atlas, one of the most heavily studied cancer datasets available, and found a stronger immune signature in invasive lobular breast cancer than had been recognized before. They then checked that result against an independent patient dataset and in lab analysis of tumor samples.
That matters because invasive lobular carcinoma has long been treated as “immune cold,” meaning not especially responsive to immunotherapy. The new work suggests that view may be too narrow. The paper, published today as “Surprisal-based large language models reveal immunologic insights in breast cancer,” points to a subtype that may deserve broader study in future immunotherapy research. The source says this could matter for roughly 15% of breast cancers diagnosed in the US each year.
Ai2 is pitching AutoDiscovery as a partner for scientists, not a replacement. It uses large language models to generate and rank surprising hypotheses, then leaves the validation to researchers and standard lab methods. That framing is doing a lot of work here, but the collaboration at least has one thing going for it: it found something unusual, and the team actually checked it.
Providence is now standing up the system inside its own cloud environment so the protected research and clinical data stay within Providence. PARC’s computational research team will install, run, and support it. Ai2 says this is an important milestone; Providence calls it a natural next step. Both are probably right, which is rare enough in AI to be mildly suspicious.
My take — AI-written commentary, not fact-checked reporting
This is the kind of AI use case that deserves attention: narrow, testable, and forced to survive validation instead of vibes. The field has had enough grand pronouncements about “transforming medicine”; a tool that helps researchers ask a better question is a much healthier claim. Also, keeping the data inside Providence instead of flinging it around for publicity is refreshingly unglamorous, which usually means it might actually work.
Read more about this at: Allen Institute (AI2)