Transforming Unimodal Silos into Computable Medicine

Healthcare has no shortage of data. The harder problem is turning fragmented clinical and biological information into a coherent patient representation that supports better research and care. Faisal Mahmood PhD explored that challenge with Tiffany Chen MD, Rowland Pettit MD, PhD, Jorge Reis-Filho MD, PhD and Celestine Schnugg.

Chen focused on tissue, where spatial relationships between cells can reveal biology that disappears when cells are studied in isolation. Combining pathology with spatial transcriptomics and proteomics may help researchers understand disease mechanisms, predict therapeutic response, and design more precise clinical trials. Pettit moved to the other end of the scale, describing foundation models that can bring together structured records, clinical text, imaging and other longitudinal data into reusable patient-level representations.

Reis-Filho connected those approaches to drug development, describing how AI can identify biomarkers that traditional methods miss and how foundation models can dramatically reduce the manual annotation required to build new tools. Schnugg brought the investor and biobanking perspective, where disease-specific translational datasets can help connect emerging models to questions that matter in real clinical development.

The larger idea was not simply to build bigger models. It was to make medicine more computable across scales, from tissue architecture to an entire patient trajectory, while preserving the ability to search, predict and explain why a model concluded.