398-AI Virtual Spatial Proteomics from Histopathology

Paper Talk · Jan 15, 2026 · 18m

Researchers have developed a deep learning framework called HEX that generates virtual spatial proteomics directly from standard, low-cost H&E-stained tissue images. While traditional spatial profiling is expensive and difficult to scale, this computational approach accurately predicts the distribution of 40 different protein biomarkers with single-cell resolution. By integrating these virtual maps with traditional pathology through a fusion model named MICA, the system significantly improves the ability to predict patient survival and immunotherapy response in lung cancer. This technology…