Local Subgroup Discovery on Attributed Network Graphs

About

This Code accompanies the paper "Local Subgroup Discovery on Attributed Network Graphs" which is part of the IDA2025 conference. The conference proceedings are published in Springers Lecture Notes in Computer Science (LNCS).

The authors are Carl Vico Heinrich, Tommie Lombarts, Jules Mallens, Luc Tortike, David Wolf, and Wouter Duivesteijn of the Eindhoven University of Technology. Corresponding author is Wouter Duivesteijn who can be reached via [email protected].

Python Environment

• The code has been tested with the Python version 3.12.9.
• The dependencies are listed in the requirements.txt file and can be run with pip install -r requirements.txt.

License

The code is licensed under MIT. Please see the LICENSE.txt file for more information.

Datasets

The code uses the following datasets:

• OGBG-MolPCBA (Primary Dataset)
• Twitch PT imported from Torch Geometric Datasets
• WebKB Cornell imported from Torch Geometric Datasets

Running the Code

Depending on the dataset, the code has to be run differently.

• For the OGBG-MolPCBA dataset, one first has to run OGBG-MolPCBA_inspecting.ipynb to inspect the dataset and to generate the .pkl file which is needed by OGBG-MolPCBA_dataset.ipynb. After that, one can run OGBG-MolPCBA_dataset.ipynb which runs the proposed algorithm on the dataset and does the ablation study. It also generates a .csv file with the results of the algorithm which is needed as input for the OGBG-MolPCBA_visualization.ipynb file which visualizes the results. Inside each of the files one has to specify the protein number which one wants to inspect, the protein number has to match with the files run prior.
• For the Twitch-PT and the WebKB Cornell dataset one first has to run the {dataset name}_dataset.ipynb file to run the proposed algorithm on the respective dataset.