Delphi-2M AI Predicts 1,000+ Diseases Up to 20 Years Before Onset

An international research team led by the University of Copenhagen has unveiled Delphi-2M, an AI model designed to forecast risk for more than 1,000 diseases, in some cases up to 20 years before onset. Researchers and commentators have likened it to a digital “crystal ball” for health, capable of mapping how diseases progress together. The work, published with collaborators including EMBL and the German Cancer Research Center, could mark a new frontier in predictive medicine as healthcare systems worldwide grapple with rising multimorbidity.

If validated, Delphi-2M could help clinicians anticipate conditions such as heart attacks, cancers, and sepsis long before symptoms emerge. By modelling “highways” of disease progression, the tool aims to improve treatment planning, avoid overtreatment, and support more efficient resource allocation. Researchers stress its potential for tackling multimorbidity—when patients develop several chronic conditions at once—which remains a major challenge in global healthcare.

• Training set: ~400,000 participants from the UK Biobank.
• Validation set: ~1.9 million patient records from the Danish National Patient Registry.
• Approach: Transformer-based generative modelling inspired by large language models, adapted to forecast disease trajectories and risk.
• Scope: More than 1,000 diseases simultaneously, rather than single-disease focus.

Professor Søren Brunak of the University of Copenhagen explained: “We wanted to explore whether it’s possible to develop a method that can handle more than 1,000 diseases simultaneously. Our study shows that it is.”

Although the research is currently at the prototype stage, the project has, according to the Mirage.News article, drawn interest from several institutions and funding bodies, including support from the Novo Nordisk Foundation. The authors indicate an intention to expand the dataset well beyond the initial 400,000 participants to further assess and refine the AI model. This collaboration between academic institutions and industry stakeholders is presented as a route that could help move the model from research prototype towards potential clinical use.

The project is still prototype-stage. Accuracy drops over longer prediction horizons, and results are stronger for diseases with consistent progression (e.g. cancers, cardiovascular disease) than for unpredictable ones (e.g. mental health, infectious disease, pregnancy-related complications). The model is not ready for clinical use, and the team acknowledges the need for rigorous trials, ethical oversight, and international regulatory review.

Funding & next steps

The project has received backing from the Novo Nordisk Foundation and other partners, with plans to expand datasets and refine predictive capacity. Future collaborations with international health agencies and additional funding are expected to pave the way for clinical trials. The ultimate goal is to integrate AI-driven risk forecasts into preventative, personalised healthcare worldwide.

Delphi-2M demonstrates the promise—and challenges—of AI in healthcare. By forecasting long-term disease risks across more than 1,000 conditions, it offers a tantalising glimpse of personalised medicine’s future. Yet, as experts caution, clinical adoption will require data expansion, independent replication, and robust safeguards to ensure both accuracy and trustworthiness.

For further updates on AI in healthcare, see the FineSkyAi Neural Network News Archive.