Pathologist-annotated tissue regions provide a fundamental reference for examining spatial omics data, yet such annotations are available for a limited number of samples due to the substantial manual effort required. Moreover, these annotations are derived from morphology within individual histology images, which can overlook molecularly defined regions and obscure intra-sample heterogeneity. To address these limitations, we present HiCAT, a machine-learning framework that automatically generates pathologist-informed region annotations and characterizes regional heterogeneity in spatial omics data. Across seven datasets, HiCAT consistently outperforms state-of-the-art methods, achieving a median relative improvement of 107% in accuracy. Beyond transferring pathologist annotations, HiCAT uncovers molecularly informed regional heterogeneity not captured by original annotations, including tumor subregions associated with clinical outcomes and brain subregions aligned with spatiotemporal disease progression. By generating consistent, highly granular, and biologically informative region annotations across large cohorts, HiCAT enables scalable downstream analysis and provides training labels for foundation models in spatial biology.
China has approved the world’s first invasive brain-computer chip—here’s what’s next
One day last October, sitting in the courtyard of his house in China’s Henan province, Dong Hui decided to see if he could hold a