DeepRIG

A deep learning-based model for gene regulatary networks (GRNs) inferrence from scRNA-seq data that transforms gene expression matrix into a correlation-based co-expression network and decouples the non-linear gene regulation patterns using graph autoencoder model (GAE).

Please see our manuscript here for more details.

Dependencies

DeepRIG is tested to work under Python 3.7. Other dependencies are list as follows:

• tensorflow 2.4
• numpy 1.19
• pandas 1.3
• scanpy 1.9
• scipy 1.7
• networkx 2.7
• scikit-learn 1.0

Installation

Option 1: Conda environment

Installing within a conda environment is recommended. After Anaconda is installed in your OS, create a new environment.

>> conda create -n new_environ_name python=3.7

The new_environ_name is the new environment name with any name you prefer. Then activate your environment using following command:

>> conda activate new_environ_name

Installing all the dependencies in this repository using conda and pip:

>> conda install -c conda-forge cudatoolkit=11.0 cudnn=8.0 -y

>> mkdir -p $CONDA_PREFIX/etc/conda/activate.d
>> cat > $CONDA_PREFIX/etc/conda/activate.d/env_vars.sh <<'EOF'
>> export LD_LIBRARY_PATH=$CONDA_PREFIX/lib:$LD_LIBRARY_PATH
>> EOF

>> pip install --upgrade pip setuptools wheel
>> pip install tensorflow==2.4.0

>> conda install -c conda-forge scanpy=1.9

Option 2: Python venv environment

Alternatively, DeepRIG can be installed in a Python virtual environment using venv. Please note that python -m venv only creates an isolated Python environment. It does not install CUDA or cuDNN. Therefore, CUDA 11.0 and cuDNN 8.0 must be available on the system, for example through HPC environment modules or an existing CUDA/cuDNN installation.

Create and activate a virtual environment:

>> python3.7 -m venv ~/venvs/deeprig_tf240
>> source ~/venvs/deeprig_tf240/bin/activate

If CUDA and cuDNN are provided by environment modules on an HPC system, load the required versions before installing or running TensorFlow:

>> module load cuda/11.0
>> module load cudnn/8.0

Alternatively, if CUDA 11.0 and cuDNN 8.0 are installed in a custom location, add their library path to LD_LIBRARY_PATH:

>> export LD_LIBRARY_PATH=/path/to/cuda_cudnn/lib:$LD_LIBRARY_PATH

Install dependencies of DeepRIG:

>> pip install tensorflow==2.4.0
>> pip install scanpy==1.9
>> pip install numpy scipy pandas scikit-learn matplotlib seaborn networkx tqdm

Usage

Inferring gene regulatary networks from scRNA-seq data

To infer gene regulatary networks from scRNA-seq data using main.py script with the following options:

• input_path string, the path of input dataset
• output_path string, the path of DeepRIG's output
• cv int, Folds for cross validation (Default 3)
• ratio int, Ratio of negative samples to positive samples (Default 1)
• dim int, The dimension of latent representations (Default 300)
• hidden1 int, Number of unites in hidden layers (Default 200)
• epochs int, Number of epochs to train (Default 500)
• learning_rate float, Initial learning rate (Default 0.01)
• dropout float, Dropout rate in all layers in GCNs (Default 0.7)

Note: The names of gene expression file and ground truth file are expected as "DatasetName" + "-ExpressionData.csv"/"-network.csv".

Example: Inferring GRNs from scRNA-seq of mouse embryonic stem cells (mESC) using DeepRIG by following codes:

>> python main.py --input_path ./Datasets/500_ChIP-seq_mESC/ --output_path ./output/ --cv 5

Outputs

• Inferred_result_dataset_name.csv Inferred gene regulation associations ranked by their edgeweights.

Evaluation

Example: To evaluate the inferred results of DeepRIG from mESC dataset, run the following command:

python evaluate.py --pred_file ./output/Inferred_result_500_ChIP-seq_mESC.csv --network ./Datasets/500_ChIP-seq_mESC/500_ChIP-seq_mESC-network.csv

Cell-type specific GRNs inferring

DeepRIG also provides the cell-type specific GRNs inference. A Jupyter Notebook of the tutorial is accessible for the cell-type specific GRNs inference.

Datasets

Demo datasets used in DeepRIG:

• hESC Human embryonic stem cells
• mESC mouse embryonic stem cells
• mDC mouse dendritic cells
• mHSC-E Erythroid lineages of mouse hematopoietic stem cells
• mHSC-L Lymphoid lineages of mouse hematopoietic stem cells
• mHSC-GM Granulocyte-macrophage lineages of mouse hematopoietic stem cells

Citing

If you find our paper and code useful, please consider citing as follows:

@article{DeepRIG,
  title={Inferring gene regulatory network from single-cell transcriptomes with graph autoencoder model},
  author={Jiacheng, Wang and Yaojia, Chen and Quan, Zou},
  journal={PLoS Genetics},
  doi = {10.1371/journal.pgen.1010942},
  year={2023},
}