The code was built based on MOSES. Thanks a lot for their code sharing!
- Python 3.6
- PyTorch 1.6
- CUDA 10.2
- RDKit
python setup.py install
Unzip file zinc_all.7z, and run code python build_dataset.py --path PATH --target TARGET to preprocess the active molecules,
where <PATH> is the path of input file, <TARGET> is the name of dataset.
To train model using MolSty, use:
python ./scripts/train.py MolSty --target TARGET --model_save ./checkpoints --n_batch 16 --n_ins 2
To generate molecules using MolSty, use:
sh generate.sh PATH_CHECKPOINT
where, <PATH_CHECKPOINT> is the checkpoint path, for example, ./checkpoints/ckpt_100.pt