This repository contains the code to replicate the experiment of CIRCE's manuscript, accessible here

Organisation

The code is a snakemake workflow, from data downloading to network inference.
The figures are then generated in the notebook Figures_notebook.

You will find here:

1. The short benchmark of the packages CIRCE, in python, and Cicero, in R.
   • This comparison is done on 2 datasets (PBMC_10X and BMMC_neurips2021) but can easily be extended to other datasets.
2. The scripts to process the human fetal scATAC atlas, used to evaluate CIRCE on very large datasets.

Execution

1. Build singularities

You will first need to install singularity/apptainer and to re-create the containers of CIRCE and Cicero:

singularity build envs/circe.sif envs/circe.def

2. Run jobs

You can now produce the output of interest for you with:

snakemake -c {number_of_cores} {output_requested} --use-singularity
# Replacing {number_of_cores} and {output_requested} by your values.

Prior to this command, you can check the jobs that will be launched to produce this output with:

snakemake -n {output_requested}

! The scripts calling Cicero might run in more than a day, hence I would advice you to run only one/few job(s) at a time.

3. Draw figure panels

You can then reproduce the figure panels which have all necessary outputs available.