Code for Benchmarking Computational Methods for Emerging Drug-Drug Interaction Prediction

Installation

1. Set Up Environment for DDI_ben, TextDDI and DDI-GPT

The models MLP, MSTE, Decagon, SSI-DDI, MRCGNN, SAGAN and TIGER in DDI_ben and TextDDI, DDI-GPT share the same environment. Our running environment is a Linux server with Ubuntu. You can set up the environment as follows:

# Create and activate Conda environment
conda create -n ddibench python=3.8.0
conda activate ddibench

# Install dependencies
# We provide the exact versions of packages we use
pip install -r DDI_ben/requirements.txt

2. Set Up Environment for EmerGNN

EmerGNN require different environments. It should be set up separately according to their respective official repositories. You can find the official repositories here:

• EmerGNN: EmerGNN Repository

Running the Code

First, cd into the corresponding directory, i.e., DDI_ben, TextDDI, EmerGNN/Drugbank, EmerGNN/TWOSIDES or SumGNN. After that,

• For DDI_ben, you can run the code as follows:

python main.py --model MSTE  --dataset drugbank --dataset_type random  --lr 3e-3 --gpu 0 

• For TextDDI,

python drugbank/main_drugbank.py --dataset_type finger --gamma_split 55

• For EmerGNN,

python -W ignore evaluate.py --dataset=S0_finger_55 --n_epoch=40 --epoch_per_test=2 --gpu=0

• For DDI-GPT,

python drugbank/main_drugbank.py --split_strategy cluster

Real Scene

Real_scene_drugbank includes DDI data from DrugBank, where drugs are divided into three sequential training-validation-test sets based on their market approval timeline.

Citation

To cite DDI-Ben in publications, please use the following BibTeX entries.

@article{shen2025benchmarking,
  title={Benchmarking drug-drug interaction prediction methods: a perspective of distribution changes},
  author={Shen, Zhenqian and Zhou, Mingyang and Zhang, Yongqi and Yao, Quanming},
  journal={Bioinformatics},
  year={2025},
  publisher={Oxford University Press}
}