Spatial gene expression at single-cell resolution from histology using deep learning with GHIST
Problem
Spatially resolved transcriptomics is costly and complex, and existing methods to predict gene expression from histology images have limited accuracy and spatial resolution.
Why it matters
Cheap, accurate in-silico spatial gene expression prediction could broaden disease-mechanism research and translational capacity.
What the AI did
GHIST, a deep learning framework, predicts spatial gene expression at single-cell resolution from histology by leveraging subcellular spatial transcriptomics and multi-layer biological information.
Results
Validated on public datasets and TCGA with claimed superior performance; specific metrics not stated in source
Impact
Peer-reviewed tool enabling in-silico single-cell spatial gene expression prediction from routine histology.
Record
| Venue | Nature Methods |
|---|---|
| Published | 2025 |
| Work type | article |
| Authors | Xiaohang Fu, Yue Cao, Beilei Bian, Chuhan Wang, J. Dinny Graham, Nirmala Pathmanathan, Ellis Patrick, Jinman Kim et al. |
| Citations | 40 (OpenAlex, at import time) |
| References | 91 |
| Model or method | GHIST (deep learning framework) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1038/s41592-025-02795-z |
| Open access | hybrid |
| Keywords | spatial transcriptomics, histology, deep learning, single-cell, gene expression prediction |
Importance rating
real but narrower advance
Cost/complexity barriers in spatial transcriptomics limit translational capacity; in-silico prediction could broaden research, but no scale or adoption evidence given.