scSGL

Graph signal processing based signed graph learning for gene regulatory inference from single cell RNA-seq data as described in [1].

Installation

Once you download the repo, go to the repo directory and start a terminal. First, create an environment and then install the required packages listed in requirements.txt. This can be done as follows for conda:

conda create -n sgl python=3.8
conda activate sgl

This will create a conda environment. To install the requirments:

conda install -c conda-forge --file requirements.txt

This may take some time as it also installs R to conda environment. Finally, to be able to use zero inflated Kendall tau as a kernel, pcaPP R package needs to be installed. This can be done by first starting R in a terminal while sgl environment is activated. Then install pcaPP by:

install.packages("pcaPP")

Usage

Please see demo.ipynb under notebooks folder for an illustration about how to use the code.

Datasets

An example dataset is given under data folder to be used in the demonstration notebook. For datasets used in the paper, please see BEELINE. Datasets used for parameter sensitivity will be published soon.

References

[1] scSGL: Signed Graph Learning for Single-Cell Gene Regulatory Network Inference