ORGANIC (Objective-Reinforced Generative Adversarial Network for Inverse-design Chemistry) is an efficient molecular generation tool, able to create molecules with desired properties. It has a user-oriented interface, and doesn't require a HPC cluster. Feel free to check our article about ORGANIC and/or contact the developers if you have any issue or are interested in collaborations.
This implementation is authored by Carlos Outeiral ([email protected]), Benjamin Sanchez-Lengeling ([email protected]), Gabriel Guimaraes ([email protected]) and Alan Aspuru-Guzik ([email protected]), affiliated to Harvard University, Department of Chemistry and Chemical Biology, at the time of release.
To install, just clone our repo:
git clone https://github.com/aspuru-guzik-group/ORGANIC.git
And, it is done!
- tensorflow==1.9
- future
- rdkit
- keras
- numpy
- scipy
- pandas
- tqdm
- pymatgen
- dill
Memo: For Mac, pip install keras-tqdm is needed.
ORGANIC has been carefully designed to be simple, while still allowing full customization of every parameter in the models. Have a look at a minimal example of our code:
model = ORGANIC('OPVs') # Loads a ORGANIC with name 'OPVs'
model.load_training_set('opv.smi') # Loads the training set (molecules encoded as SMILES)
model.set_training_program(['PCE'], [50]) # Sets the training program as 50 epochs with the PCE metric
model.load_metrics() # Loads all the metrics
model.train() # Proceeds with the trainingIf you are interested, we encourage you to read our '20 minutes to ORGANIC' tutorial, where you can learn all (literally, all) the functionalities of ORGANIC.