Skip to content

AI for discovering 100% valid drug like molecules, a combination of VAE-JTNN and bayesian optimization, an optimized Python 3 Version of Junction Tree Variational Autoencoder for Molecular Graph Generation (ICML 2018)

License

Notifications You must be signed in to change notification settings

cosmina98/FastJTNNpy3

 
 

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

21 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

FastJTNNpy3 : Junction Tree Variational Autoencoder for Molecular Graph Generation

Python 3 Version of Fast Junction Tree Variational Autoencoder for Molecular Graph Generation (ICML 2018)

Implementation of our Junction Tree Variational Autoencoder https://arxiv.org/abs/1802.04364

Requirements

  • RDKit (version >= 2017.09) : Tested on 2019.09.1
  • Python (version >= 3.6) : Tested on 3.7.4
  • PyTorch (version >= 0.2) : Tested on 1.0.1

To install RDKit, please follow the instructions here https://www.rdkit.org/docs/Install.html

We highly recommend you to use conda for package management.

Quick Start

Code for Accelerated Training

This repository contains the Python 3 implementation of the new Fast Junction Tree Variational Autoencoder code.

  • fast_molvae/ contains codes for VAE training. Please refer to fast_molvae/README.md for details.
  • fast_jtnn/ contains codes for model implementation.
  • fast_bo/ contains codes for Bayesian Optimisation (WIP: support for custom rdkit functions).
  • fast_molopt/ contains codes for molecule optimisation using a JTpropVAE which is the same as JTVAE but also enmeds properties with the molecules. (WIP: integration in main pipeline)

Old codes

This repository contains the following directories:

  • Old/bo includes scripts for Bayesian optimization experiments. Please read Old/bo/README.md for details.
  • Old/molvae/ includes scripts for training our VAE model only. Please read Old/molvae/README.md for training our VAE model.
  • Old/molopt/ includes scripts for jointly training our VAE and property predictors. Please read Old/molopt/README.md for details.
  • Old/molvae/jtnn/ contains codes for model formulation. Please read Old/molvae/README.md for training our VAE model.

Contact

Bibhash Chandra Mitra ([email protected])

About

AI for discovering 100% valid drug like molecules, a combination of VAE-JTNN and bayesian optimization, an optimized Python 3 Version of Junction Tree Variational Autoencoder for Molecular Graph Generation (ICML 2018)

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • Jupyter Notebook 72.1%
  • Python 27.9%