ARAE for molecular generation with SMILES representation

pytorch version (recommand)

https://github.com/gicsaw/ARAE_torch

Prerequisite:

python3

numpy

tensorflow <=1.13

RDKit

SA_Score for RDKit

https://github.com/rdkit/rdkit/tree/master/Contrib/SA_Score

Download:

git clone https://github.com/gicsaw/ARAE_SMILES

Training data preparation :

cd ARAE_SMILES

For ZINC

python data_char_ZINC.py train

python data_char_ZINC.py test

For QM9

python data_char_QM9.py train

python data_char_QM9.py test

training:

ARAE with QM9 dataset

python train_ARAE_QM9.py

ARAE with ZINC dataset

python train_ARAE_ZINC.py

CARAE with ZINC dataset for logP, SAS, and TPSA

python train_CARAE_logP_SAS_TPSA.py

Test:

We prepared trained hyperparameters in save dir.

Test for ARAE with QM9

python test_n_ARAE_QM9.py

Test for ARAE with ZINC

python test_n_ARAE_ZINC.py

Test for CARAE with ZINC (conditional)

python test_n_CARAE_con_logP_SAS_TPSA.py $logP $SAS $TPSA

Test for CARAE with ZINC (unconditional)

python test_n_CARAE_uncon_logP_SAS_TPSA.py

Molecular generation:

Molecular generation for ARAE with QM9

python gen_ARAE_QM9.py

Molecular generation for ARAE with ZINC

python del_end_code.py out_ARAE_QM9/79

generated smiles: out_ARAE_QM9/79/smiles_gen.txt

python gen_ARAE_ZINC.py

python del_end_code.py out_ARAE_ZINC/39

generated smiles: out_ARAE_ZINC/39/smiles_gen.txt

Molecular generation for CARAE with ZINC (conditional)

python gen_CARAE_con_logP_SAS_TPSA.py $logP $SAS $TPSA

References:

Hong, S. H., Ryu, S., Lim, J., & Kim, W. Y. (2019). Molecular Generative Model Based On Adversarially Regularized Autoencoder. Journal of Chemical Information and Modeling.