NIDDK Will Fund Exactly One $5 Million Multimodal AI Team for Type 1 Diabetes: RFA-DK-28-116 Lands February 10, 2027

Academic PIs who build multimodal AI for chronic disease have a new NIDDK target: RFA-DK-28-116, a single $5 million U01 for type 1 diabetes precision medicine, forecast on Grants.gov on August 20, 2026, with applications estimated due February 10, 2027.

That is the whole opportunity in one sentence, and almost every word of it is unusual. One award. One team. A named artificial intelligence mandate from an institute that has spent two decades building the datasets the winner will be asked to integrate. And a receipt date that sits far enough from NIH's standard February council cycle to be easy to miss.

What NIDDK Actually Forecast on August 20

The record lives on Grants.gov as opportunity 363678, posted August 20, 2026 by the National Institutes of Health on behalf of the National Institute of Diabetes and Digestive and Kidney Diseases. The title is "Multimodal AI to Accelerate Precision Medicine for Type 1 Diabetes (U01 – Clinical Trials Not Allowed)."

The forecast's numbers are specific in the way forecasts usually are not:

• Estimated program funding: $5,000,000
• Expected number of awards: 1
• Funding instrument: Cooperative agreement (U01)
• Cost sharing: Not required
• Estimated full NOFO posting: December 1, 2026
• Estimated application due date: February 10, 2027
• Estimated award date: November 1, 2027
• Estimated project start: December 1, 2027
• Fiscal year: 2028
• Assistance listing: 93.847, Diabetes, Digestive, and Kidney Diseases Extramural Research
• Contact: NIDDK Division of Diabetes, Endocrinology and Metabolic Diseases, NIDDK_DEM@nih.gov

A single expected award against a $5 million estimate is the tell. This is not a portfolio-building RFA that will fund six R01-scale efforts and see what sticks. NIDDK is describing one coordinated project with one set of milestones, and the review will be structured accordingly — comparative, winner-take-all, and heavily weighted toward whichever team can credibly claim it already has the data access and the engineering bench.

Eligibility is broad. The forecast's applicant-type list includes public and private institutions of higher education, nonprofits with and without 501(c)(3) status, state and local governments, tribal governments and organizations, small businesses, and other for-profits. It also explicitly names non-domestic entities — foreign institutions are eligible, which is not a given on U-mechanism awards and matters for the European and Australian groups sitting on complementary islet and registry data.

The Datasets Are the Real Prize

Read NIDDK's own framing and the strategy becomes clear. The forecast description spends its first paragraph not on AI but on inventory: over the past two decades NIDDK has funded major T1D consortia across basic, translational, and clinical research, each focused on a different organ, disease stage, or clinical outcome, and each generating "large longitudinal and multimodal datasets using standardized protocols and coordinated studies."

Anyone who works in this space can fill in the names — the Human Islet Research Network and its Human Pancreas Analysis Program, the Network for Pancreatic Organ donors with Diabetes, TEDDY, TrialNet, the NIDDK Central Repository, the Integrated Islet Distribution Program, dkNET. Individually, each has produced a coherent picture of one slice of the disease. Collectively, they have never been modeled as a single system.

NIDDK says so directly: combining the consortia's work "offers a rare opportunity to see how different biological processes interact to drive T1D development and progression," and that integrated view "can answer fundamental questions about T1D biology that cannot be answered using individual datasets or studies alone."

The deliverables the forecast names are worth quoting because they are unusually concrete for a pre-NOFO document. The project will develop and test advanced multimodal AI models, and it will also develop AI-enabled research workflows, new methods to integrate data, and privacy-preserving synthetic datasets. That last item is the one to underline. Synthetic data generation is being written into the scope as a first-class deliverable, not a compliance afterthought — which suggests NIDDK expects the resulting models and derived datasets to circulate beyond the funded team, and that the consent and governance friction across donor-tissue and clinical-trial cohorts is a known blocker the award is meant to solve.

The scientific aims follow from that: new insight into T1D biology, better resolution of the disease's different forms and stages, biomarker identification, and new approaches to prevention, intervention, and treatment. Endotype discovery, in other words. The field has spent years arguing that "type 1 diabetes" is several diseases wearing one name; this award is NIDDK putting $5 million behind the computational case.

"Clinical Trials Not Allowed" Is a Scoping Instruction

The parenthetical in the title does real work. It means no interventional aim can appear anywhere in the application, and NIH's clinical-trial questionnaire will be the gate. Teams that instinctively bolt a small pilot intervention onto the back of a computational project to demonstrate translational reach need to resist that reflex here — it converts a competitive application into a withdrawn one.

It also clarifies what the $5 million is buying. This is a data-and-methods award: compute, engineering, harmonization, model development, validation against held-out consortium data. Budget accordingly, and expect reviewers to look skeptically at any line item that reads like patient recruitment.

The U01 designation carries the second constraint. A cooperative agreement means substantial NIH programmatic involvement — NIDDK program staff will sit inside the project, and the forecast's phrase "coordinated, milestone-driven research project" is not decoration. Expect the eventual NOFO to require a milestone table with go/no-go criteria, annual review against those milestones, and continuation contingent on hitting them. Applicants who have only ever run R01s should build that table early; it is the single most common place where otherwise strong U01 applications lose points.

Where This Sits Against NIH's Other AI Money

Across NIH's forecasted opportunities, only about a dozen currently carry an explicit artificial intelligence framing, and RFA-DK-28-116 is the only one aimed at a single disease with a dedicated institute-level dataset inheritance behind it. That non-overlap is the strategic point.

The obvious comparison is the Common Fund's imaging-and-multimodal program, which we covered in detail in our breakdown of NIH's PRIMED-AI five-RFA structure. PRIMED-AI is broad, trans-NIH, imaging-anchored, and built to fund a network. RFA-DK-28-116 is narrow, single-institute, tissue-and-registry-anchored, and built to fund one team. A group with real multimodal AI capability can plausibly write to both without triggering scientific overlap — different data, different aims, different councils — and the sequencing helps, since the T1D full NOFO is estimated for December 1, 2026, well after PRIMED-AI's cycle.

The deadline geography matters too. February 10, 2027 is a distinct receipt date, not a standard NIH cycle date. It will not appear in your institution's default deadline calendar, and your sponsored programs office will not surface it automatically. Put it in manually.

The Fifteen Weeks Before the NOFO Drops

Everything above is a forecast. Dates, dollar figures, and award counts can and do move between forecast and published NOFO, and NIDDK is explicit that these are estimates. But a forecast this specific — named mechanism, named institute division, dollar figure, single award, four dated milestones — is a strong signal that the program is through internal approval and the concept is settled.

That gives roughly fifteen weeks before December 1 to do the work that cannot be done in the eight weeks between NOFO and deadline:

Secure data access now. Every consortium named above has its own application, DUA, and review timeline. A team that shows up in February 2027 with executed agreements in hand is a different applicant than one describing an intention to apply for access. Start with the NIDDK Central Repository and PANC-DB.

Write the milestone table before you write the aims. Cooperative agreements are evaluated on whether the project can be managed, not just whether it is interesting.

Assemble the harmonization case. The scientific novelty is integration. Reviewers will want to see that you have confronted the actual mismatches — assay platforms, staging definitions, longitudinal sampling cadence, consent scope — rather than asserted that a foundation model will absorb them.

Contact program staff. NIDDK_DEM@nih.gov is the listed contact on the forecast. Pre-NOFO conversations are the cheapest signal available about scope and fit, and with one expected award, fit is everything.

Track the mechanism and the field around it: search active NIH multimodal AI and precision medicine solicitations on Granted to see what else is open in this space while you wait for December 1.