CellGRN

Introduction

CellGRN is a framework that shifts the paradigm from global fitting to cell-specific reconstruction. CellGRN adopts a two-step strategy: it first leverages a global regulatory backbone from a global modeling strategy, and subsequently reconstructs cell-specific regulatory weights by projecting individual cell states onto this backbone.

CellGRN-sparse is an accelerated version of CellGRN, which skipped the step of cell-specific GRN output and optimized with sparse matrix and block computation.

All datasets can be downloaded from FigShare or baidu Net Dist(https://pan.baidu.com/s/1mkpBOzyyuOHvmTPC4sMGBg?pwd=4r8v)

Overview of CellGRN and CellGRN-sparse

---

Pakcage dependencies

• Python >=3.8, scanpy, scikit-learn,pyranges,umap-learn, anndata,pandas, numpy, scipy, jupyter
• Visualization requires extra matplotlib, seaborn and R>=3.5.

Installation

If you have not installed python dependencies above, we recommend to use conda in install CellGRN in an independent environment. The conda tool (miniconda) can be installed from anaconda website.

• Step1: create a conda environment.

conda env create -f cellgrn_env.yml -n cellgrn
conda activate cellgrn

2. Install cellgrn package.

git clone https://github.com/xxx/cellgrn
# set dir to folder
cd cellgrn
pip install .

3. Test the installation in python

import cellgrn

Tutorial

A demo code for CellGRN and CellGRN-sparse tutorial

Repository structure

• code/grn_method: SCENIC+ and LINGER script.
• code/preprocess: Preprocessing scripts for input datasets and validation datasets.
• code/***_run.ipynb: Demo code to run CellGRN in multiome/spatial-multiome datasets in our manuscript.
• eval/***_eval.ipynb: Benchmark pipelines for each dataset.
• eval/visualization/: R scripts to plot figure in our manuscript.

Reference

Reconstructing cell-specific GRN to decipher regulatory heterogeneity from single-cell multi-omics. (submited)