Learning Graph-Level Representation for Drug Discovery

Paper Link: Learning Graph-Level Representation for Drug Discovery

Requirements

• Install DeepChem(july2017)

Usage

1.Clone the repository

git clone https://github.com/ZJULearning/graph_level_drug_discovery.git

2.Training

python train.py --gpu 0 --dataset pcba

Our train.py only supports 6 datasets in MoleculeNet, including Tox21, ToxCast, HIV, MUV, PCBA, SAMPL.

Result

Database and baseline: MoleculeNet

Dataset	Split Method	Train	Valid	Test
Tox21	Index	0.965	0.839	0.848
Tox21	Random	0.964	0.842	0.854
Tox21	Scaffold	0.971	0.788	0.759
ToxCast	Index	0.927	0.747	0.734
ToxCast	Random	0.924	0.746	0.768
ToxCast	Scaffold	0.929	0.696	0.657
PCBA	Index	0.904	0.869	0.864
PCBA	Random	0.899	0.863	0.867
PCBA	Scaffold	0.907	0.847	0.845

Citation

Please cite our work in your publications if it helps your research:

@article{Li2017Learning,
  Title={Learning Graph-Level Representation for Drug Discoveryk},
  Journal={arXiv preprint arXiv:1709.03741},
  Author={Junying Li, Deng Cai, Xiaofei He},
  Year={2017},
}