• python anaconda3(installed rdkit, chainer)

  • pubchempy (for converting PunChem ID into smiles)

  • node2vec, networkx(for applying node2vec algorithm)

  • pyensembl (for converting ensembl protein ID into amino acid sequences)

  pyensembl install --release 93 --species human

1. Make input file for integrate model Hard dataset already processed is contained in dataset_hard except for 'protein.npy'(protein onehot). Easy dataset and middle dataset is saved as input files (csv format) that haven't preprocessed yet in dataset_easy and dataset_middle.

1.1 run preprocessed_dataset.py

python preprocessed_dataset.py --input <INPUT_FILE>

<INPUT_FILE> input file (only csv format is available)

2. Apply node2vec to multi-interactome data

node2vec model of chemical-chemical interaction data are already saved as data_multi/modelcc.pickle. node2vec model of protein-protein interaction data are already saved as data_multi/modelpp.pickle.

construct graph network connecting compounds(proteins) interacting each other and apply the graph to node2vec here.

2.1 run omics_compound.py

python omics_compound.py

chemical_chemical_interaction.csv (multi-interactome data) is setteing as default input file.

2.2 run omics_protein.py

python omics_protein.py

protein_protein_interaction.csv (multi-interactome data) is setteing as default input file.

3. Learn and predict by integrated model

You can demonstrate our program with test data.

cd /path/to/multiDTI/data_multi
python omics_compound.py
python omics_protein.py

• 5 fold Cross-validation for hard-dataset

  • PreProcessed train dataset

  python preprocessed_dataset.py --input <FILEPATH> --data <DATA>

  <FILEPATH> './dataset_hard'

  <DATA> './train'

  • PreProcessed test dataset

  python preprocessed_dataset.py --input <FILEPATH> --data <DATA>

  <FILEPATH> './dataset_hard'

  <DATA> './test'

  • Learn

  cd /path/to/multiDTI
  python training.py --input <FILEPATH> --output <DIRECTORY>

  • Predict

  python evaluation.py --input <FILEPATH> --output <DIRECTORY> --place <PLACE_DIRECTORY>

• Watanabe, N., Ohnuki, Y., Sakakibara, Y.
  • Deep learning integration of molecular and interactome data for protein-compound interaction prediction