A Self-Attention Based Message Passing Neural Network for Identifying Structure-Property Relationships

Introduction

This is a PyTorch implementation of the research: A Self-Attention Based Message Passing Neural Network for Identifying Structure-Property Relationships

Environment

Python 3.6.5
Pytorch 1.0 
RDkit 2018.03.4 
Autograd 1.2 
Numpy 1.14.2 
Pandas 0.23.4 
tqdm 3.7.1

Data File

Data file format:
Datafile can be CSV and text as showd in the data_RE2.

train

python reg_wat.py #replace the data_path and cols_to_read as your want
The trained model weights will be also stored in save_test.

prediction and visualization

python viz_wat.py #replace the data_path and checkpoint_path as your want
The results of prediction will be also stored in save_test and visualization file will be save in png_*.

repeat this work

bash go_repeat.sh

Acknowledgement

We thank the previous work by Ramsundar and swansonk14 teams. The code in this repository is inspired on DeepChem and ChemProp

nanomolar / SAMPN