A pathologist stains a tissue section with hematoxylin and eosin — cells turn blue and pink, and under the microscope you can see their shape and arrangement. But which genes are active in each cell cannot be read from such an image. That requires a separate, expensive assay that reveals the activity of thousands of genes at once, but it is rarely performed.

STORM looks at a stained image and predicts which genes are turned on at each spot. It does so more accurately than other methods across 11 tumor types. The model works with different platforms (Visium, Xenium, Visium HD, CosMx) and requires no retuning. On data from 7245 patients across 23 independent cohorts, STORM predicted whether a patient would respond to immunotherapy and how the disease would progress better than standard biomarkers.

Such a tool turns a routine histological image into a molecular map. A doctor sees not only the shape of cells but also their internal program, and can choose treatment more precisely.