This code reproduces the experimental results obtained with the AMCG model as presented in the paper
AMCG: a graph dual atomic-molecular conditional molecular generator
C. Abate, S. Decherchi, A. Cavalli
To install the conda environment used to run the scripts just run
conda env create -f environment.yaml
Download the data from here, extract the .zip file and copy the content of AMCG_DATA in the data folder.
Activate the environment via conda activate amcg_env
To train a model from scratch create a training configuration file and run
python train.py path/to/config/file
All the possible options can be found in configs/train_config_guide.ini.
To sample new molecules create a sampling configuration file and run
python sample.py path/to/config/file
All the possible options can be found in configs/sample_config_guide.ini.
To optimize existing molecules create an optimization configuration file and run
python optimize.py path/to/config/file
All the possible options can be found in configs/optim_config_guide.ini.
Working configuration files can be found in configs folder.
The notebook eval.ipynb contains the code used to generate tables and figures.
AMCG is GPL-licensed. Please see license.txt file in the repository.