Generative adversarial networks for molecules:
This repository contains all the code necessary to train a GAN to generate new molecules using keras/tensorflow gpu.
The dataset used to train the model was downloaded from the enamine website: https://enamine.net/compound-libraries/diversity-libraries/dds-50240. The dataset contains a total of 50,240 molecules.
Use the SmilesToImage function to convert a csv file containing smiles to images with a resolution of 300X300 pixels. Requires installation of RdKit.
Use the ImageToSmiles function to make a csv datafrane containing smiles for images of molecules in a folder. Requires installation of decimer.
ImageToSmiles and SmilesToImage support use of multiple processors for quick conversion.
- Clone/Download the repository.
- Install dependencies using conda env install -f environment.yml.