Objectives & Ambition
AIRIS aims to create a new generation of Generative AI for biomedical research, combining multimodal data with causal and mechanistic knowledge to produce more robust, interpretable and biologically meaningful insights. Moving beyond today’s largely correlation-based AI, AIRIS strives to support researchers throughout the discovery process, from data integration to evidence-based hypothesis generation and study design.
- 01
Harmonising and Processing Multimodal Datasets
Providing agentic automated and semi-automated pipelines for harmonising and combining diverse data, including omics, imaging, clinical records, patient-reported outcomes and lifestyle information, while using synthetic data to address gaps, under-representation and privacy constraints – creating analysis-ready data.
- 02
Moving from Correlations to Mechanisms
AIRIS will move beyond current Generative AI models, which are limited by correlation-based reasoning, hallucinations, and lack of interpretability. Embedding biological knowledge, causal relationships and dynamic models into Generative AI to better represent disease processes across molecular, cellular, tissue, organ, and patient levels.
- 03
Developing an Active Scientific Collaborator
Developing an AI-assisted research infrastructure that can synthesise data and scientific knowledge, generate and prioritise novel hypotheses, and support the design of rigorous studies to test them.
- 04
Building Trustworthy and Usable AI
Establishing transparent evaluation and benchmarking approaches for accuracy, robustness, fairness, interpretability and usability, while integrating ethical, legal and societal considerations throughout development. AIRIS will set a new standard for evaluating biomedical Generative AI.
- 05
Creating an Open and Sustainable Foundation
Developing scalable, reproducible and accessible AI tools and infrastructure, leveraging European computing and research ecosystems and making results openly available wherever possible to support long-term uptake by research, healthcare and industry.
- 06
Demonstrating Real-World Scientific Value
Validating the approach across five diverse disease areas – Pulmonary Fibrosis, Steatotic Liver Disease, Cardiovascular Disease, Chronic Kidney Disease and Inflammatory Bowel Disease – to demonstrate generalisability and potential value for predictive and personalised medicine.