Genrative Adversarial Networks for De Novo Molecular Design

This repository contains the original PyTorch implementation of the paper 'Generative Adversarial Networks for De Novo Molecular Design'.

A. Requirements

Anaconda (>= 4.8.3)
PyTorch (tested on 1.7.0)
PyTorch Lightning (tested on 1.1.2)
Tensorboard (tested on 2.4.0)

B. Dataset

Donwlonad from the following link: https://www.ebi.ac.uk/chembl/
Or change the ChEMBL version and data save format to csv from the following link https://github.com/BenevolentAI/guacamol/blob/master/guacamol/data/get_data.py and run.
Use make_data.ipynb to create the data to be used for SMILES-MaskGAN and save it as a pickle.

C. Reference Code

Refer to the https://github.com/jerinphilip/MaskGAN.pytorch link for the PyTorch-based MaskGAN source code. Fairseq: This code uses the package provided at the https://github.com/pytorch/fairseq link.

D. SMILES-MaskGAN training

Command line: python run_maskgan.py