CosGeneGate Selects Multi-functional and Credible Biomarkers for Single-cell Analysis

This is the official repo for our marker gene selection project.

Installation

To prepare the model, please use:

pip install -r requirements.txt

To install stg, please use our forked mode from this link:

git clone https://github.com/HelloWorldLTY/stg.git
cd stg/python
pip install -e ./

To install the R environment for the deconvolution analysis, please use:

install.packages("CosGeneGateDeconv")

or

install.packages('devtools')
devtools::install_github('VivLon/CosGeneGate/tree/main/CosGeneGateDeconv')
library(CosGeneGateDeconv)

Tutorials

Please use the file Tutorial_marker_selection.ipynb as a tutorial for running our algorithm. The demo dataset can be found in this folder.

Benchmark

Please refer the original repos for the implementation of our benchmarked methods: COSG, NS-Forest, scGeneFit, STG, and scMAGS.

Acknowledgements

We refer the codes from STG, CSCORE, and COSG.

Thanks for their great work!

Citation

@article{liu2025cosgenegate,
  title={CosGeneGate selects multi-functional and credible biomarkers for single-cell analysis},
  author={Liu, Tianyu and Long, Wenxin and Cao, Zhiyuan and Wang, Yuge and He, Chuan Hua and Zhang, Le and Strittmatter, Stephen M and Zhao, Hongyu},
  journal={Briefings in Bioinformatics},
  volume={26},
  number={1},
  pages={bbae626},
  year={2025},
  publisher={Oxford University Press}
}