Spatial biomarker discovery via interpretable semantic learning in histopathology

PathPrism is a new computational framework that converts whole-slide histopathology images into interpretable spatial biomarker spectra, providing a transparent representation of tissue architecture. The system enables high-performance linear modeling of prognosis, molecular alterations, and treatment response across cancer types. By replacing opaque deep-learning predictions with interpretable spatial features, the approach transforms digital pathology into a platform for hypothesis-driven biomarker discovery and perturbation-driven exploration.