This repository is the python implementation of paper 'GraphIX: Graph-based In silico drug repoisioning with XAI'.
XaiDR is interpretable supervised learning framework for drug repositioning. It can present important genes that have high contribution to disease-drug association prediction.

GraphIX is testet to work with python 3.6. The required dependencies are:

networkx
numpy
pandas
tensorflow>=1.12
joblib
scipy
scikit-learn>=0.21
rdkit

In the top directory, run the code as following:

sh preprocessing/generate_kg.sh

python gcn.py train_cv --config config.json

As an example, calculate the contribution of surrounding 1-hop nodes to Intestinal neplasms(node number: 15958)-Salicylic acid(node number: 20380) novel edge mentioned in the paper.

python gcn.py --config config.json visualize --visualize_type edge_score --visualize_node0 20380 --visualize_node1 15958 --graph_distance 1

@article{takagi2022,
  title={GraphIX: Graph-based In silico XAI for drug repositioning from biopharmaceutical network},
  author={Atsuko Takagi, Mayumi Kamada, Eri Hamatani, Ryosuke Kojima, Yasushi Okuno},
  doi = {https://doi.org/10.48550/arXiv.2212.10788},
  month = {12},
  year={2022}
}