python 3.7
tensorflow-gpu >= 2.3.0
rdkit >= 2020.03.2.0
pandas >= 1.0.3
numpy >= 1.19.2

See data fold to get data and preprocess data.

python training.py --input_npy ./data/training_raw_chain --target_npy ./data/training_raw_target --npy_num 10
More arguments setting can be found and changed in code file training.py.

python eval_random.py --num_sample NUM_SAMPLE --output OUTPUT --ckpt CHECKPOINT
More arguments setting can be found and changed in code file eval_random.py.

python eval_from_scaffold.py --input INPUT_NPY --target TARGET_NPY --num_sample NUM_SAMPLE --output OUTPUT_FILE --ckpt CHECKPOINT_PATH
More arguments setting can be found and changed in code file eval_from_scaffold.py.

Zhirui Liao, Lei Xie, Hiroshi Mamitsuka, Shanfeng Zhu, Sc2Mol: a scaffold-based two-step molecule generator with variational autoencoder and transformer, Bioinformatics, Volume 39, Issue 1, January 2023, btac814, https://doi.org/10.1093/bioinformatics/btac814

Please feel free to send bug reports or questions to Zhirui Liao: [email protected] and Prof. Shanfeng Zhu: [email protected]

It is free for non-commercial use. For commercial use, please contact Zhirui Liao and Prof. Shanfeng Zhu ([email protected]).