Accurate and interpretable drug-drug interaction prediction enabled by knowledge subgraph learning

Contents

• Overview
• Repo Contents
• System Requirements
• Installation Guide
• Examples
• Results
• License
• Issues
• Citation

Overview

Background: Discovering potential drug-drug interactions (DDIs) is a long-standing challenge in clinical treatments and drug developments. Recently, deep learning techniques have been developed for DDI prediction. However, they generally require a huge number of samples, while known DDIs are rare.

Methods: In this work, we present KnowDDI, a graph neural network-based method that addresses the above challenge. KnowDDI enhances drug representations by adaptively leveraging rich neighborhood information from large biomedical knowledge graphs. Then, it learns a knowledge subgraph for each drug-pair to interpret the predicted DDI, where each of the edges is associated with a connection strength indicating the importance of a known DDI or resembling strength between a drug-pair whose connection is unknown. Thus, the lack of DDIs is implicitly compensated by the enriched drug representations and propagated drug similarities.

Results: Here we show the evaluation results of KnowDDI on two benchmark DDI datasets. Results show that KnowDDI obtains the state-of-the-art prediction performance with better interpretability. We also find that KnowDDI suffers less than existing works given a sparser knowledge graph. This indicates that the propagated drug similarities play a more important role in compensating for the lack of DDIs when the drug representations are less enriched.

Conclusions: KnowDDI nicely combines the efficiency of deep learning techniques and the rich prior knowledge in biomedical knowledge graphs. As an original open-source tool, KnowDDI can help detect possible interactions in a broad range of relevant interaction prediction tasks, such as protein-protein interactions, drug-target interactions and diseasegene interactions, eventually promoting the development of biomedicine and healthcare.

Repo Contents

• data: the pre-processed dataset of Drugbank and BioSNAP.
• pytorch: the pytorch version code of KnowDDI.
• raw_data: the origin dataset of Drugbank and BioSNAP.

System Requirements

Hardware Requirements

This repository requires only a standard computer with enough RAM to support the in-memory operations. We recommend that your computer contains a GPU.

Software Requirements

OS Requirements

The package development version is tested on Linux(Ubuntu 18.04) operating systems with CUDA 10.2.

Python Dependencies

For the pytorch version, the environment required by the code is as follows.

python==3.7.15
pytorch==1.6.0
torchvision==0.7.0
cudatoolkit==10.2
lmdb==0.98
networkx==2.4
scikit-learn==0.22.1
tqdm==4.43.0
dgl-cu102==0.6.1

Installation Guide

For the pytorch version, please follow the commands below:

git clone git@github.com:tata1661/KnowDDI-codes.git
cd KnowDDI-codes
conda create -n KnowDDI_pytorch python=3.7
conda install pytorch==1.6.0 torchvision==0.7.0 cudatoolkit=10.2 -c pytorch
pip install dgl-cu102==0.6.1
pip install -r requirements.txt
cd pytorch

Examples

The default parameters are the best on Drugbank dataset. To train and evaluate the model,you could run the following command.

python train.py -e Drugbank

Besides, to train and evaluate the model on BioSNAP dataset,you could run the following command.

python train.py -e BioSNAP --dataset=BioSNAP --eval_every_iter=452 --weight_decay_rate=0.00001 --threshold=0.1 --lamda=0.5 --num_infer_layers=1 --num_dig_layers=3 --gsl_rel_emb_dim=24 --MLP_hidden_dim=24 --MLP_num_layers=3 --MLP_dropout=0.2

Dataset

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).

We provide the mapping file between ids in our pre-processed data and their original name/drugbank id as well as a copy of hetionet data and their mapping file on this link.

Results

We provide examples on two datasets with expected experimental results and running times.

Citation

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

@article{wang2023accurate,
  title={Accurate and interpretable drug-drug interaction prediction enabled by knowledge subgraph learning},
  author={Wang, Yaqing and Yang, Zaifei and Yao, Quanming},
  journal={arXiv preprint arXiv:2311.15056},
  year={2023}
}

Acknowledgement

The code framework is based on SumGNN.