This repository is the source code of our paper "PHGL-DDI: A pre-training based hierarchical graph learning framework for drug-drug interaction prediction".
python=3.7.12
pytorch-1.11.0
cuda-11.3
torch_cluster-1.6.0
torch_scatter-2.0.9
torch_sparse-0.6.13
torch_spline_conv-1.2.1
torch-geometric= 2.0.4
GPU: NVIDIA GeForce RTX 4090 GPUs were used for model training. Each GPU has 24GB of memory, which accelerates large-scale matrix operations and deep learning model training.
CPU: The system is equipped with an Intel 13900KF processor, featuring 24 cores to support data preprocessing and other computational tasks.
Memory: The system memory is configured with 64GB to ensure there are no bottlenecks when handling large-scale datasets.
PubChem dataset ：contains15.56 million unlabeled molecules，each molecule is represented by SMILES.
DrugBank dataset : contains 191,808 DDI tuples with 1706 drugs, each drug is represented in SMILES.
ChCh-Miner dataset : contains 959 drugs and 33,669 DDIs, with each row of data containing the IDs of the two drugs and their corresponding SMILES.
Zhang dataset
[1] Gao Z, Jiang C, Zhang J, et al. Hierarchical graph learning for protein-protein interaction. Nat Commun. 2023;14(1):1093. Published 2023 Feb 25. doi:10.1038/s41467-023-36736-1.
[2] Wang, Y., Wang, J., Cao, Z. et al. Molecular contrastive learning of representations via graph neural networks. Nat Mach Intell4, 279–287 (2022).