Charlotte Bunne (ETHZ 2026)
Cancer is among the most complex diseases in medicine. A single patient can generate dozens of different measurements, from tissue images and genomic sequences to molecular maps and clinical records. Each captures a different aspect of the disease, yet clinicians must integrate them to decide on the best treatment. At the same time, another challenge is the counterfactual: we observe what happens along the treatment path that is chosen, but never what might have happened under a different therapy, at a different dose, or at a different time.
The “Virtual Patient Labs” project aims to change this by building a computational model of each patient’s disease that draws all available data into a single, coherent picture. First, a multimodal foundation model learns to represent a patient’s biology across molecular, cellular, and tissue scales. Second, a generative simulator forecasts how the disease might evolve under different treatments, making those unseen alternatives computationally accessible. Third, an AI reasoning system translates these predictions into concrete clinical support: recommending therapies, selecting the most informative diagnostic tests, and identifying biological markers that explain why a treatment succeeds or fails. In collaboration with the University Hospitals of Geneva and Zurich, the platform will be evaluated on retrospective patient cohorts across several cancer types.