Source for paper DGNN-DDI is designed to a dual GNN for drug-drug interaction prediction based on molecular structure and can provide explanations that are consistent with pharmacologists.

We have added comments to drugbank/data_preprocessing.py and drugbank/model.py. If you are interested in the technical details of preprocessing steps and algorithms, I think those comments would be helpful.

numpy ==1.22.3
pandas == 1.4.3
python == 3.8.13
pytorch == 1.12.0
rdkit == 2020.09.1
scikit-learn ==1.1.1
torch-geometric ==2.0.4
torch-scatter == 2.0.9
torch-sparse == 0.6.14
tqdm == 4.64.0

• First, run data_preprocessing.py using ' data_preprocessing.py -d drugbank -o all`
  Running data_preprocessing.py convert the raw data into graph format.

• Second, run train.py using ' train.py --fold 0 --save_model'