Codes and data for "Employing Molecular Conformations for Ligand-based Virtual Screening with Equivariant Graph Neural Network and Deep Multiple Instance Learning"
If you make advantage of the EquiVS model and our benchmark dataset proposed in our paper, please consider citing the following in your manuscript:
Gu, Y.; Li, J.; Kang, H.; Zhang, B.; Zheng, S. Employing Molecular Conformations for Ligand-Based Virtual Screening with Equivariant Graph Neural Network and Deep Multiple Instance Learning. Molecules 2023, 28, 5982. https://doi.org/10.3390/molecules28165982
Download our released dataset files from Google Drive. Then unzip the data folder to this repo. The files are as shown:
data
molecules
The molecular conformer SDF files.target
The sub-datasets (CSV files) for specific targets, columns include 'Activity type', 'Ligand name', 'SMILES', 'Conformer_path', etc.molecule_structure.csv
The mapping file for each molecule SMILES and its corresponding conformer path in 'data' folder.
Our bioactivity prediction dataset is curated from "A consensus compound/bioactivity dataset for data-driven drug design and chemogenomics"(https://zenodo.org/record/6398019#.YwuT93ZByUk). The data preprocessing cases can be found in data_preprocessing.ipynb.
- torch==1.8.0
- dgl==0.5.2
Simply reproduce our model:
sh bash.sh
Note that the TARGET_processed folder will be created in data/task/ to save molecular graphs (dgl_graph.bin) and labels (info.pkl).
OR:
python main.py -id {DEVICE_ID} -sp {SAVE_PATH} -se {SEED} -mo {MODE: train OR eval}
For more (general, training, and model) arguments, please see main.py.
The EquiVS model will automatically train on all stored sub-dataset in data/target folder. The training results, models, and predictions will be stored in result/{SAVE_PATH} folder.
Please refer the model interpretation case in model_interpret.ipynb.
We welcome you to contact us (email: yg3191@nyu.edu) for any questions and cooperations.