The article introduces a novel deep learning framework named GHIST (Gene expression from HISTology) designed to predict spatially resolved gene expression at a single-cell resolution using routinely collected H&E-stained histology images. Recognizing the high cost and complexity of current spatial transcriptomics technologies, GHIST leverages multitask deep learning to integrate multiple biological information layers, including cell type, nuclei morphology, and neighborhood composition, to enhance prediction accuracy. The authors demonstrate that GHIST outperforms existing spot-based methods and can enrich large datasets like The Cancer Genome Atlas (TCGA) with spatial omics data, thereby facilitating advanced multi-omics analysis and the discovery of new disease insights and biomarkers. The research validates GHIST's performance using breast cancer, lung adenocarcinoma, and melanoma datasets, highlighting its utility in survival prediction and uncovering the effects of genomic alterations on spatial gene patterning.

References:

• Fu X, Cao Y, Bian B, et al. Spatial gene expression at single-cell resolution from histology using deep learning with ghist[J]. Nature Methods, 2025: 1-11.