sCellST is a novel method for inferring gene expression from H&E images trained on spatial transcriptomics data.

The code has been tested with python 3.10 and cuda 11.8. To install the envrionment we used poetry which you can install from poetry. Make sure to use a version at least >= 2.0. Then, you can simply run your code with poetry run

git clone https://github.com/loicchadoutaud/sCellST.git
cd sCellST
poetry install
poetry add --editable ./external/cell_SSL   # For SSL training
poetry run python full_pipeline_script.py

The code now used the HEST database (https://github.com/mahmoodlab/HEST) as raw data since it allows to use many publicly released spatial transcriptomics.

The code is structured as follows:

• scellst: source code for the sCellST method + benchmarked methods in scellst/bench
• external: external code used in the project
• sCellST_reproducibility/reproducibility_figures contains all the notebooks used to produce the figures from the paper
• sCellST_reproducibility/submit_scripts scripts contains all the scripts used to submit the jobs on the cluster with submitit (https://github.com/facebookincubator/submitit)
• simulation: contains all the code for the simulation experiments

We provide a notebook to run the full pipeline on a breast cancer slide from the HEST database in training_tutorial.ipynb (takes between 10 and 15 minutes).

# Reference scRNA-seq dataset
# 
wget https://datasets.cellxgene.cziscience.com/73fbcec3-f602-4e13-a400-a76ff91c7488.h5ad -O data/raw_ovary_dataset.h5ad
https://www.nature.com/articles/s41588-021-00911-1
wget https://datasets.cellxgene.cziscience.com/fabd4946-3f41-459c-ba79-188749a8baa4.h5ad -O data/raw_breast_dataset.h5ad

https://www.nature.com/articles/s41467-025-67965-1

@article{chadoutaud_scellst_2026,
	title = {{sCellST} predicts single-cell gene expression from {H}\& {E} images},
	url = {https://www.nature.com/articles/s41467-025-67965-1},
	doi = {10.1038/s41467-025-67965-1},
	journal = {Nature Communications},
	author = {Chadoutaud, Loïc and Lerousseau, Marvin and Herrero-Saboya, Daniel and Ostermaier, Julian and Fontugne, Jacqueline and Barillot, Emmanuel and Walter, Thomas},
	year = {2026},
}

We thanks the authors of the HEST database (https://github.com/mahmoodlab/HEST) and the original Mocov3 (https://github.com/facebookresearch/moco-v3) adapted for this project