You can carry out the GxRNN method with the following steps.

1. Make environment by gxrnn.yml.

Execute the following command on terminal. 

conda env create --file gxrnn.yml
conda activate gxrnn

If you can't expand yml file, please install some libraries written in env/requirements.txt.

2. Train GxRNN.

Execute the following command on terminal.

python main.py --train_gxrnn --smiles_epochs 500 --hidden_size 1024 --smiles_dropout 0.3

3. Generate molecules from target-perturbed profiles.

Execute the following command on terminal.

python main.py --hidden_size 1024 --generation --protein_name {protein_name (ex.AKT1)}

4. Calculate Tanimoto coefficients between newly generated molecules and known ligands.

Execute the following command on terminal.

python main.py --calculate_tanimoto --candidate_num 1000 --protein_name {protein_name (ex.AKT1)}

To generate molecules with a pretrained model and reproduce the results in the article, execute the following command.

python main.py --hidden_size 1024 --generation --calculate_tanimoto --saved_gxrnn results/saved_gxrnn_hidden1024_dropout03_500.pkl --candidate_num 1000 --protein_name {protein_name (ex.AKT1)}

If you have any questions, please ask Yuki Matsukiyo.
Email: matsukiyo.yuki566@mail.kyutech.jp
