HetDDI

Overview

This repository is the source code of our paper "HetDDI: a pre-trained heterogeneous graph neural network model for drug-drug interaction prediction".

Environment Setting

This code is based on Pytorch and dgl-cuda. You need prepare your virtual enviroment early.

Running the code

You can run the following command to run our work:

python main.py

There are several parameters can be customized:

1. batch_size
2. label_type, you can choose one of "multi_class", "binary_class" or "multi_label".
   "multi_class" and "binary_class" is only available at ddi_name = "DrugBank".
   "multi_label" is only available at ddi_name = "TWOSIDES"
3. condition, you can choose one of scenarios s1, s2, s3 defined in our paper, default is "s1"
4. mode, you can choose one of variants "HetDDI-mol", "HetDDI-kg" or "HetDDI" by "only_mol", "only_kg", "concat". Default is "HetDDI" by "concat".
5. ddi_name, the dataset you want to run, "DrugBank" or "TWOSIDES". Default is "DrugBank"

Dataset Preparation

The dataset used in paper is available at /HetDDI/data/DRKG+DrugBank and /HetDDI/data/DRKG+TWOSIDES

If you want to use yourself dataset, you need to follow these format.

1. nodes.tsv

the form is look like:

1618 Compound::DB09499 0

• node id
• node name
• node type id

2. edges.tsv

the form is look like:

318 14 30460

• head node id
• relation id
• tail node id

3. smiles.tsv

the form is look like:

9 CN+(C)CCO

• node(drug) id
• smiles string

4. ddi.tsv

the form is look like:

78 1616 59

• node(drug) id
• node(drug) id
• interaction type

Weight files

The weight files for the model can be obtained from the following link.

The path for the weight files should be the root directory of project.

https://drive.google.com/drive/folders/1VKbVVzAcv_e3UgxId-Jrpac2SKqnCWeN