CSGNN

The code for paper "CSGNN: Contrastive Self-Supervised Graph Neural Network for Molecular Interaction Prediction"

1. Overview

The repository is organized as follows:

• data/ contains the datasets used in the paper;
• parms_setting.py contains hyperparameters adopted by CSGNN;
• data_preprocess.py is the preprocess of data before training;
• layer.py contains mix-hop GNN layers and contrastive GNN layers;
• instantiation.py instantiates the CSGNN;
• train.py contains the training and testing code on datasets;
• utils.py contains preprocessing functions of the data (e.g., normalize...);
• main.py contains entry to CSGNN (e.g., normalize...);

2. Dependencies

• numpy == 1.18.5
• scipy == 1.5.2
• sklearn == 0.23.2
• torch == 1.5.0
• torch-geometric == 1.6.1
• networkx == 2.4

3. Example

Here we provide several example of using CSGNN: To run CSGNN with GCN decoder on DTI network using "uniform" as initial features and output the result to test.txt, execute the following command:

python main.py --aggregator GCN --feature_type uniform --in_file data/DTI.edgelist --out_file test.txt

To run CSGNN with GIN decoder on DDI network using "position" as initial features and output the result to test.txt, execute the following command:

python main.py --aggregator GIN --feature_type position --in_file data/DDI.edgelist --out_file test.txt