Developing a Mechanism-Informed AI Collaborator for Biomedical Research

Use Cases

To demonstrate the applicability and added value of AIRIS, the AI collaborator will be evaluated across five complementary biomedical use cases using large-scale datasets. These span different organs, medical fields, data types and clinical needs, while also sharing important biological mechanisms and risk factors. Applying a common approach across all five disease areas will allow AIRIS to assess its scientific value and generalisability across different biomedical contexts. Validation against current state-of-the-art methods will use predefined measures of predictive performance, robustness, interpretability and usability, helping determine which approaches may be transferable beyond individual use cases.

A Common Approach Across All Use Cases

The same core methodologies will be applied across the five disease areas:

01

Multimodal data onboarding and automated harmonisation.

02

Mechanism-informed model development, grounding AI outputs in biological plausibility.

03

Hypothesis generation and evaluation using the AIRIS muti-agent hypothesis generation system.

04

Proof-of-concept validation against current state-of-the-art methods, using pre-defined evaluation metrics for predictive performance, robustness, interpretability, and usability.

Use Cases in Detail

Explore the five use cases below and select one to learn more about each of them.

Pulmonary Fibrosis

Pulmonary Fibrosis is a group of progressive lung disorders characterised by irreversible scarring that can ultimately lead to respiratory failure. In Idiopathic Pulmonary Fibrosis (IPF), median survival after diagnosis is only three to five years. Disease progression varies considerably between patients, and although antifibrotic therapies can slow progression, there is currently no cure.

By integrating clinical, imaging and molecular data, AIRIS will support predictive models and clinically relevant hypotheses to better understand disease progression and treatment response. This could contribute to patient-specific digital twins and more individualised predictions of disease progression and therapeutic response. Generated hypotheses will be assessed for plausibility and novelty and complemented by biological cross-validation, supporting a proof of concept for AI-enabled precision medicine in pulmonary fibrosis.

Expert teams evaluating AIRIS usability in Pulmonary Fibrosis:Yale, KU Leuven, ICS-HUB, CHUV, UNIBO, AUSL Romagna

Pulmonary Fibrosis