Disease subnetwork detection with explainable Graph Neural Networks

Paper

The paper for this project is available here: https://academic.oup.com/bioinformatics/article/38/Supplement_2/ii120/6702000

A readthedocs documentation of GNN-SubNet is in progress and can be found here: https://gnn-subnet.readthedocs.io

See also: https://github.com/pievos101/Ensemble-GNN/blob/main/README.md for disease module discovery using ensemble learning with Graph Neural Networks

Installation

To install GNNSubNet run:

pip install torch 
pip install torch-geometric
pip install torch-scatter
pip install torch-sparse

pip install GNNSubNet
# newest version from GitHub can be installed via source
pip install GNN-SubNet/

Preferred versions are: torch==1.11.0, torchvision==0.12.0, torch-geometric==2.0.4, torch-scatter==2.0.9, and torch-sparse==0.6.13.

Usage

Synthetic Barabasi Networks

The datasets can be found here: https://github.com/pievos101/GNN-SubNet/tree/main/GNNSubNet/datasets/synthetic

from GNNSubNet import GNNSubNet as gnn

# Synthetic data set  ------------------------- #
loc   = "/home/bastian/GitHub/GNN-SubNet/GNNSubNet/datasets/synthetic"
ppi   = f'{loc}/NETWORK_synthetic.txt'
feats = [f'{loc}/FEATURES_synthetic.txt']
targ  = f'{loc}/TARGET_synthetic.txt'

# Read in the synthetic data
g = gnn.GNNSubNet(loc, ppi, feats, targ, normalize=False)

# Get some general information about the data dimension
g.summary()

# Train the GNN classifier and validate performance on a test set
g.train()

# Check the performance of the classifier
g.accuracy
g.confusion_matrix

# Run the Explainer with 4 iterations (10 is recommended)
g.explain(3)

# Edge and Node (Gene) Importances 
g.gene_names
g.edges

# Importances
g.edge_mask
g.node_mask

# Detected modules and their importances
g.modules
g.modules[0]

g.module_importances

TCGA multi-omics kidney cancer

The datasets can be found here: https://github.com/pievos101/GNN-SubNet/tree/main/TCGA

from GNNSubNet import GNNSubNet as gnn

# location of the files
loc   = "/home/bastian/GitHub/GNN-SubNet/TCGA"
# PPI network
ppi   = f'{loc}/KIDNEY_RANDOM_PPI.txt'
# single-omic features
#feats = [f'{loc}/KIDNEY_RANDOM_Methy_FEATURES.txt']
# multi-omic features
feats = [f'{loc}/KIDNEY_RANDOM_mRNA_FEATURES.txt', f'{loc}/KIDNEY_RANDOM_Methy_FEATURES.txt']
# outcome class
targ  = f'{loc}/KIDNEY_RANDOM_TARGET.txt'

# Load the multi-omics data 
g = gnn.GNNSubNet(loc, ppi, feats, targ)

# Train the GNN classifier and validate performance on a test set
g.train()

# Check the performance of the classifier
g.accuracy
g.confusion_matrix

# Run the Explainer with 4 iterations (10 is recommended)
g.explain(3)

# Edge and Node (Gene) Importances 
g.gene_names
g.edges

# Importances
g.edge_mask
g.node_mask

# Detected modules and their importances
g.modules
g.modules[0]
g.gene_names[g.modules[0]]

g.module_importances

Additional informations

The GNNSubNet initialization function expects the PPI network as an input, the feature matrices, as well as the outcome class. The input needs to be adjusted by the user.

The PPI network consists of three columns. The first two columns reflect the edges between the nodes (gene names), the third column is the confidence score of the specified edge. The range of this score is [0,999], where high scores mean high confidence.

The rows of the feature matrices (e.g mRNA, and DNA Methylation) reflect the patients, whereas the columns represent the features (genes). Row names as well as column names are required!

Please see the folder "datasets/TCGA" for some sample/example files.

The mentioned workflow performs GNN classification, explanations, and community detection for disease subnetwork discovery.

After exectution, importance scores are stored in the 'edge_mask.txt' file of the data folder.

The detected disease subnetworks can be found within the 'communities.txt' file, and the corresponding scores within the 'communities_scores.txt' file.

Note, the mentioned community file contain the network node identifier which match the order of the gene names stored in 'gene_names.txt'

Citation

https://academic.oup.com/bioinformatics/article/38/Supplement_2/ii120/6702000

Bibtex

@article{pfeifer2022gnn,
  title={{GNN-SubNet}: Disease subnetwork detection with explainable graph neural networks},
  author={Pfeifer, Bastian and Saranti, Anna and Holzinger, Andreas},
  journal={Bioinformatics},
  volume={38},
  number={Supplement\_2},
  pages={ii120--ii126},
  year={2022},
  publisher={Oxford University Press}
}