A world model of patient biology

Most “virtual cell” and protein-level foundation models in biology operate at the wrong level of abstraction; drugs act on tissues and patients, not on isolated cells. We are building the first multi-modal biological foundation model designed to reason about neurological disease at the patient level. Trained on a decade of proprietary patient data, our model links brain biology, longitudinal biofluids, and clinical trajectory into a unified representation of each patient, and reasons about the molecular state and clinical progression of these patients and how they might respond to a given treatment.

The model incorporates brain biology from tissue, blood biology from living patients, and bridges the two using a smaller set of paired samples. It is being trained to read out molecular brain state from more accessible modalities like biofluids, imaging, and clinical attributes, allowing scientists and clinicians to retrieve what we call a virtual biopsy of the brain. This unlocks the ability for the field to directly interrogate their hypothesized mechanism of action of a treatment directly in the brain as soon as patients receive the first dose — a feat never before achievable.

From there, it is designed to match patients to the right therapies, predict how patients will evolve over the course of their disease, and predict how patients will respond to different interventions as reflected in both peripheral fluids (biomarkers) and the brain (disease state).