This is the code for our paper ``SumGNN: Multi-typed Drug Interaction Prediction via Efficient Knowledge Graph Summarization'' (published in Bioinformatics'21) [link].


git clone [email protected]:yueyu1030/SumGNN.git
cd SumGNN
pip install -r requirements.txt


    -d drugbank         # task
    -e ddi_hop3         # the name for the log for experiments
    --gpu=0             # ID of GPU
    --hop=3             # size of the hops for subgraph
    --batch=256         # batch size for samples
    --emb_dim=32        # size of embedding for GNN layers
    -b=10               # size of basis for relation kernel

You can also change the d to BioSNAP. Please change the e accordingly. The trained model and the logs are stored in experiments folder. Note that to ensure a fair comparison, we test all models on the same negative triplets.


We provide the dataset in the data folder.

Data Source Description
Drugbank This link A drug-drug interaction network betweeen 1,709 drugs with 136,351 interactions.
TWOSIDES This link A drug-drug interaction network betweeen 645 drugs with 46221 interactions.
Hetionet This link The knowledge graph containing 33,765 nodes out of 11 types (e.g., gene, disease, pathway,molecular function and etc.) with 1,690,693 edges from 23 relation types after preprocessing (To ensure no information leakage, we remove all the overlapping edges between HetioNet and the dataset).

Knowledge Graph Embedding

We train the knowledge graph embedding based on the framework in OpenKE.

To obtain the embedding on your own, you need to first feed the triples in train.txt (edges in dataset) and relations_2hop.txt (edges in KG) as edges into their toolkit and obtain the embeddings for each node. Then, you can incorporate this embedding into our framework by modifying the line 44-45
in model/dgl/

Cite Us

Please kindly cite this paper if you find it useful for your research. Thanks!

  title={Sumgnn: Multi-typed drug interaction prediction via efficient knowledge graph summarization},
  author={Yu, Yue and Huang, Kexin and Zhang, Chao and Glass, Lucas M and Sun, Jimeng and Xiao, Cao},


The code framework is based on GraIL.