ISNN

Official implementation of the paper:
"Implicit Subgraph Neural Network"
Yongjian Zhong, Liao Zhu, Hieu Vu, Bijaya Adhikari
Accepted at the 42nd International Conference on Machine Learning (ICML 2025), Poster [OpenReview PDF]

Setup

1. Install Required Libraries

To begin using the ISNN framework, ensure all necessary libraries are installed. You can install the required libraries by using the requirements.txt file. To install the dependencies, run the following command:

pip install -r requirements.txt

or use your preferred package manager.

This repository was run and tested with Python 3.8.10.

2. Datasets

The Real-World Datasets used in this project includes HPO_METAB, HPO_NEURO, PPI_BP, and EM_USER, all of which were sourced from the paper:

"Subgraph Neural Networks" Emily Alsentzer, Samuel G. Finlayson, Michelle M. Li, Marinka Zitnik
Proceedings of Neural Information Processing Systems (NeurIPS), 2020 [arXiv]

and can be downloaded from this Dropbox link here.

Once installed, unzip the contents of the folder and set the DATASET_PATH variable in the config_path.py file to the path of the folder containing the datasets.

3. Plotting Results

Inside the root (ISNN) directory, create a directory called plots with a subdirectory for each dataset you are working with. For example, to create the directories for the HPO_METAB dataset, run the following commands:

cd ~/PATH/TO/ISNN
mkdir plots
cd plots
mkdir hpo_metab

Once this is setup, you can proceed to Running the ISNN framework.

Running

To run ISNN, make sure you are cd'd into the ISNN directory and run the following command:

python ISNN.py --dataset {dataset_name} --model isnn --repeat 10

The hyperparameters used to produce the results in the paper are preset and can be found in the /hyperparams/{dataset}.yml file. To change the hyperparameters, you can modify the .yml file.

Results

A summary of the results will be written to {dataset_name}_{model}_results.json in the root directory of the project.