What if routine blood tests could do more than report today's numbers and help predict what comes next – for instance detect people at-risk of Crohn's disease at an early stage?

In our paper for the leading machine learning conference ICLR 2026, we introduce SuperMAN (Super Mixing Additive Networks) – an interpretable AI model that learns from real-world clinical measurement data, including irregular and fragmented blood tests, collected at different times and frequencies.

SuperMAN predicts outcomes such as Crohn's disease onset with state-of-the-art accuracy. It also reveals the biology behind the prediction by highlighting the biomarkers that drive risk and when critical transitions may be unfolding.

This way, SuperMAN offers something proposed clinical AI models often lack — not just a prediction, but an understanding of the biology driving it, which is essential for clinical adoption.

Authors

Andrea Zerio
Maya Bechler-Speicher
Maor Huri
Marie Vibeke Vestergaard
Ran Gilad-Bachrach
Tine Jess
Samir Bhatt
Aleksejs Sazonovs