What it is
Researchers present HEX, a model that computationally generates spatial proteomics from standard hematoxylin and eosin slides, trained and validated on 819,000 histopathology image tiles with matched protein expression from 382 tumor samples. It predicts the expression of 40 biomarkers spanning immune, structural and functional programs. Applied across six independent non-small-cell lung cancer cohorts totaling 2,298 patients, HEX-enabled multimodal integration improved prognostic accuracy by 22% and immunotherapy response prediction by 24-39% compared with conventional clinicopathological and molecular biomarkers.
Why it matters
Spatial proteomics maps protein expression at high resolution, but its cost, complexity and scalability keep it out of routine clinical use. Deriving the expression of 40 biomarkers directly from the H&E slides already produced in routine pathology lowers that barrier.
How to read this
It offers a low-cost, scalable route to interpretable spatial biomarkers, extending precision oncology from specialized assays toward routine histopathology.
Underlined numbers link to their source. See all metrics and figures under Sources below.