This README file contains the introduction to the repo "GRN_Inference", including short descriptions of the other scripts.

Introduction:

This repo contains instructions and scripts to generate data, analyze, and plot as mentioned in our paper "Gene regulation network inference using k-nearest neighbor-based mutual information estimation- Revisiting an old DREAM".
The GRN inference pipeline implemented here is modular, one can use specific functions to calculate MI quantities based on kNN and integrate the output matrix into a different inference algorithm than the ones implemented here.

Prerequisite - Cloning the repo, creating a python virtual environment and installing dependencies:

# Clone the project
git clone https://github.com/XiaoLabJHU/GRN_Inference.git

# Go into the project folder
cd GRN_Inference

# Create a virtual Python installation in the .venv folder.
python3 -m venv .venv

# Activating a virtual environment will put the virtual environment-specific python and pip
# executables into your shell’s PATH.
# For Windows OS see: https://docs.python.org/3/library/venv.html
source .venv/bin/activate

# Confirm you are using the python environment
which python

# Tell pip to install all packages in the requirements file (for specific python project)
python3 -m pip install -r requirements.txt

# Leaving the virtual environment when done
deactivate

Simulating/generating gene expression data:

The software GeneNetWeaver used to generate the datasets in the current study is available in the GitHub repository, https://github.com/tschaffter/genenetweaver
We used our Jupyter notebook Generate_replicates_for_network_inference.ipynb to call GeneNetWeaver to generate multiple replicates of steady-state and time-series gene expression datasets for realistic in silico networks of sizes of 50, and 100 genes containing various experimental conditions (knockouts, knockdowns, multifactorial perturbation, etc.).
We then combine all the steady-state expression data to a single file: {network name}_all.tsv

The data we used for the paper can also be downloaded from the DATA folder in this repo.

Density estimation and MI calculation:

The following python modules: Mutual_Info_based_binning_module.py, Mutual_Info_KNN_nats_module.py, Building_MI_matrices.py are responsible to discretize continuous data (i.e. gene expression profiles) to fixed width bins or organize the data according to k-NN. Following by mutual information calculations by one of four estimators ((i) Maximum Likelihood (ML, given by Shannon), (ii) Miller-Madow correction (MM), (iii) Kozachenko-Leonenko (KL), and (iv) Kraskov-Stoögbauer-Grassberger (KSG).

To test the accuracy of each MI estimator, we use the Python script Mutual_Information_estimators_comparisson_for_Gaussian_distribution.py to generate random variables (of different sizes) drawn from a joint Gaussian distribution (with different correlations) and calculate the various 2d and 3d MI quantities. Afterwards the Jupyter notebook CMIAwKSG_paper_Figure2_and_SI.ipynb compares the calculated results with the analytical solution.

To test the computational costs of each MI estimator, we use the script Computation_time_vs_data_size.py to generate a representative gene expression data matrix for a network of size 50 (nodes or genes) with up to 1000 data points (conditions/perturbations/time-series) and calculate the time to build 2d and 3d MI matrices. Afterwards the Jupyter notebook CMIAwKSG_paper_Figure6_computation_cost.ipynb plots the results.

GRN inference (including density estimation and MI calculation):

The script GRN_inference_and_AUPR_calc.py together with the python modules: Mutual_Info_based_binning_module.py, Mutual_Info_KNN_nats_module.py, Building_MI_matrices.py, Inference_algo_module.py, Precision_Recall_module.py, is responsible to discretize the continous gene expression data to fixed width bins or organize the data according to k-NN. Following by mutual information calculations by one of four estimators ((i) Maximum Likelihood (ML, given by Shannon), (ii) Miller-Madow correction (MM), (iii) Kozachenko-Leonenko (KL), and (iv) Kraskov-Stoögbauer-Grassberger (KSG). Finally, the network structure is infered by one of six inference algorithms (Relevance Networks, RL; Algorithm for the Reconstruction of Accurate Cellular Networks, ARACNE; Context-Likelihood-Relatedness, CLR; Synergy-Augmented CLR, SA-CLR; Luo et al. MI3; CMI2rt, and our Conditional-Mutual-Information-Augmentation, CMIA)

GRN performance evaluation and plotting:

To compare the performence and plot it we use the Jupyter notebook CMIAwKSG_paper_AUPR_figures_and_SI.ipynb.
Data is organized in a pandas dataframe and plotted using matplotlib.