Code for the paper: ChemGAN challenge for drug discovery: can AI reproduce natural chemical diversity? arXiv preprint arXiv:1708.08227.
Related blog post: https://medium.com/the-ai-lab/chemgan-challenge-for-drug-discovery-can-ai-reproduce-natural-chemical-diversity-8f1f2528ee22
Chat room: https://gitter.im/Startcrowd/drugdiscovery
Video by Siraj Raval: https://youtu.be/hY9Bc3mtphs
Requirements: Rdkit version 2017.03.3 from Anaconda, Tensorflow 1.0.1
The code has not been cleaned, don't hesitate to post an issue if you don't find what you are looking for.
To make the DRD2 case work, take clf.pkl here, rename it clf_drd2.pkl, and put it in the appropriate folder. It's the SVM activity model of DRD2 by Marcus Olivecrona.
To make the QED case work, you need Silicos-it.
In order to train the model, cd into model and run
python train_ogan.py exp.json
where exp.json is a experiment configuration file.