Seminal
The magazine and atlas of EU-funded researchA new publication from the web and AI team at Scott Charnley

Atlas / Germany

European Research Council grant · Running

ORACLE Out-of-Distribution-aware Reasoning Agents for Clinical Learning and Exploration

Coordinator
Friedrich-Alexander-Universitaet Erlangen-Nuernberg, Germany
Runs
1 January 2027 to 30 June 2028
Status
Running
EU funding
€150k
Total cost
—
Partners
1 organisations in 1 countries
Type
European Research Council grant
Grant agreement
101326871

What the project is doing

In the project's own words, from its summary on CORDIS.

We have recently observed an unexpected and profoundly promising synergy: multi-modal out-of-distribution detection methods, originally developed for normative modelling in the ERC MIA-NORMAL project, appear to endow AI agents with remarkable autonomy in real-world scientific reasoning. By explicitly encoding uncertainty and recognising unfamiliar inputs, these agents become capable of detecting anomalies, and they begin to act independently: curating data, identifying patterns, and proposing hypotheses in complex clinical domains.

This breakthrough reframes the role of AI in medicine. No longer confined to fixed-task classification or retrospective association, agentic AI systems can now act as autonomous collaborators in medical discovery. We validated this concept in cardiovascularimaging, where agents independently conducted phenome-wide association studies, rediscovered known correlations, and surfaced novel phenotype-disease links with transparent, explainable logic.

With ORACLE, we aim to extend this paradigm to oncology, focusing on pancreatic cancer, a devastating disease with limited early detection and poor outcomes. Agents will begin with large-scale, multimodal clinical data: radiology, pathology, molecular reports, and labs, and autonomously curate and structure it into patient trajectories. Once organised, agents will explore the data landscape:clustering phenotypes, linking imaging to molecular patterns, and testing hypotheses across cohorts.

When normative data is available, they compare against population-wide baselines; when it is not, agents can use our generative methods to impute missing values and identify structure. Running entirely on real clinical data within GDPR-compliant infrastructure, ORACLE will demonstrate that autonomous AI agents can produce publishable, clinically relevant insights. This PoC may not only advance cancer research but reshape how biomedical discovery is done: transforming it into an automated, data-driven process.

Who is involved

OrganisationCityCountryRole
Friedrich-Alexander-Universitaet Erlangen-NuernbergErlangenGermanyCoordinator

Source: CORDIS, European Commission. Data of 5 Nov 2026, reused under the Commission's reuse policy. View on CORDIS · Grant DOI