A novel molecular graph heterogeneous graph neural network (HGNN) for capturing local intermolecular reactivity.

This repository includes the codes and results for our paper: Capturing inter-molecular reactivity using heterogeneous graph neural networks

ReacHGNN aims to provide a general framework for solving inter-molecular reaction performance prediction tasks, including the construction method of reaction heterogeneous graph (RHG) representation and the corresponding neural network architecture (ReacHGNN). It supports the prediction of different types of reactivity indicators such as reaction yield, transition state energy barrier, enantioselectivity and reaction rate constant which are included in this repo.

Please first clone our repo and install using the setup.py. All the dependencies are listed in the requirements.txt.

git clone https://github.com/Masker-Li/ReacHGNN.git
cd ReacHGNN
conda install --yes --file requirements.txt (if needed)

The descriptions of files in each folder are listed in the corresponding README.md file in the folder

• datasets contains the raw data for four benchmark datasets and the reorganize Mayr dataset obtained from the original publications and Mayr’s Database of Reactivity Parameters.

• examples contains the notebooks and selected initial experiments used to collect all the results, and also the results presented in the manuscript.

• reachgnn contains the source codes of RHG and ReacHGNN

Questions about this repository may be addressed to Xin Li ( maskerli [AT] tencent [DOT] com ).