AIRIS - Mechanism-Informed Multimodal Generative AI for Causal and Dynamical Modelling in Biomedical Research
Project duration
06/2026 - 05/2030
Project partners
- Athina-Erevnitiko Kentro Kainotomias Stis Technologies Tis Pliroforias, Ton Epikoinonion Kai Tis Gnosis (ATHENA)
- Forschungszentrum Jülich GmbH (FZJ)
- Universiteit van Amsterdam (UVA)
- Yale University (Yale)
- The Research Institute of the McGill University Health Centre (McGill)
- The University of Manchester (UOM)
- Linköping University (LiU)
- University College London (UCL)
- Harokopio University of Athens (HUA)
- Universität Bern (UBERN)
- Uniklinik RWTH Aachen (UKA)
- Katholieke Universiteit Leuven (KU Leuven)
- Institut Català de la Salut (ICS-HUB)
- Technische Universität München (TUM)
- University of Bologna (UNIBO)
- Azienda Usl Della Romagna (AUSL Romagna)
- EURICE European Research and Project Office GmbH (EURICE)
- Elsevier (ELSEVIER)
- MLCommons Association (MLCommons)
- A. Persidis & Sia OE (Biovista)
- Timelex (TLX)
- Centre Hopitalier Universitaire Voisdois (CHUV)
Funding
The project is funded by the European Union Horizon Europe Programme - Grant Agreement Number 101289094 under the HORIZON-HLTH-2025-01-TOOL-03 topic of the HORIZON-HLTH-2025-01 call.
Project description
AIRIS will develop the next generation of multimodal Generative AI (GenAI) models to accelerate research on predictive and personalised medicine. Building on recent advances in LLMs, causal inference and mechanistic modelling, AIRIS will integrate heterogeneous biomedical data (omics, imaging, clinical, laboratory, lifestyle, PROMs and scientific literature) into mechanisminformed generative frameworks. These models will not only generate biologically plausible synthetic data to address sparsity and bias but also embed causal and dynamical constraints across biological scales, from “virtual cells” to multi-organ models, enabling counterfactual reasoning and in-silico hypothesis testing. The project will deliver (i) robust agentic multimodal data integration pipelines; (ii) novel mechanism-anchored generative architectures with interpretability and bias-mitigation safeguards; (iii) hypothesis generation assistants for biomarker discovery, drug repurposing, and intervention design; and (iv) a scalable MLOps infrastructure ensuring reproducibility. AIRIS will be validated using already existing large-scale datasets in five high-impact use cases including Pulmonary Fibrosis, Steatotic Liver Disease, Cardiovascular Disease, Chronic Kidney Disease and Inflammatory Bowel Disease. AIRIS will demonstrate utility in modelling of disease, while also allowing cross-disease insights. A multidisciplinary consortium spanning leading universities and research centres, HPC experts, industry, and SME ensures excellence across the project dimensions. ELSI and fairness will be systematically addressed by SSH and legal experts. Outputs will be openly disseminated through benchmarks, training, and open-source releases, fostering European leadership in mechanism-informed GenAI. AIRIS will maximise scientific, clinical, and societal impact by bridging AI innovation with biomedical discovery, improving disease understanding, and supporting equitable personalised care.
The SDL Fluids & Solids Engineering of JSC is leading the Work Package (WP) 4, which is on "Scalable Infrastructures, MLOPs, and Deployments". This WP is cross-sectional to the technical and use-case WPs of WP1, WP2, WP3, and WP5. In AIRIS, the SDL is responsible for managing HPC and data infrastructure access, AI-model deployment (baseline and full demonstrator stack) as well as advancing the MLOPs tool itwinai, originally developed by Forschungszentrum Jülich and CERN in the interTwin project, with features for agentic AI access and cross-site workflow synchronization.
Project website
The website of the project can be found under https://airis-ai.eu.