Skip to content
Template-free prediction of organic reaction outcomes
Jupyter Notebook Python JavaScript CSS
Branch: master
Clone or download
Latest commit cca11ec Feb 22, 2019
Type Name Latest commit message Commit time
Failed to load latest commit information.
human Human benchmarking reactions Sep 10, 2018
notebooks Fixing lingering import issues Nov 3, 2018
rexgen_direct Updates to documentation Nov 3, 2018
template_comparison Template comparison info, also used in NIPS Jun 13, 2018
website Fixing lingering import issues Nov 3, 2018
.gitignore DSStore Jun 13, 2018
LICENSE Initial commit Jun 13, 2018 Update Feb 22, 2019


Template-free prediction of organic reaction outcomes using graph convolutional neural networks

Described in A graph-convolutional neural network model for the prediction of chemical reactivity


  • Python (trained/tested using 2.7.6, visualization/deployment compatible with 3.6.1)
  • Numpy (trained/tested using 1.12.0, visualization/deployment compatible with 1.14.0)
  • Tensorflow (trained/tested using 1.3.0, visualization/deployment compatible with 1.6.0)
  • RDKit (trained/tested using 2017.09.1, visualization/deployment compatible with 2017.09.3)
  • Django (visualization compatible with 2.0.6)

note: there may be some issues with relative imports when using Python 2 now; this should be easy to resolve by removing the periods preceding package names


Looking at predictions from the test set

cd into the website folder and start the Django app using python runserver. Go to http://localhost:8000/visualize in a browser to use the interactive visualization tool

Using the trained models

You can use the fully trained model to predict outcomes by following the example at the end of rexgen_direct/rank_diff_wln/

Retraining the models

Look at the two text files in rexgen_direct/core_wln_global/notes.txt and rexgen_direct/rank_diff_wln/notes.txt for the exact commands used for training, validation, and testing. You will have to unarchive the data files after cloning this repo.

You can’t perform that action at this time.