txie-93 / polymernet

Multi-task polymer network

This repository implements the multi-task graph neural networks used in the paper "Accelerating amorphous polymer electrolyte screening by learning to reduce errors in molecular dynamics simulated properties".

Prerequisites

This software package requires:

• PyTorch
• RDKit
• PyTorch Geometric

Intructions for installing the prerequisites can be found in above websites.

Usage

We provide two different types of training procedules.

Single Task

To perform single task training, run the following command:

python single_task_train.py --log10 0 data/logp/noise_0.64

The code will perform single task training that can reduce the random errors in the noisy training data.

Multi Task

To perform multi task training, run the following command:

python multi_task_train.py data/conductivity/5ns data/conductivity/50ns

The code will perform multi task training that can reduce systematic errors between 5ns and 50ns simulations.

Rerunning all experiments

We provide a script to quickly rerun most experiments in paper.

bash run.sh

Citation

Please consider citing the following paper if you find our code & data useful.

@article{xie2022accelerating,
  title={Accelerating amorphous polymer electrolyte screening by learning to reduce errors in molecular dynamics simulated properties},
  author={Xie, Tian and France-Lanord, Arthur and Wang, Yanming and Lopez, Jeffrey and Stolberg, Michael A and Hill, Megan and Leverick, Graham Michael and Gomez-Bombarelli, Rafael and Johnson, Jeremiah A and Shao-Horn, Yang and others},
  journal={Nature communications},
  volume={13},
  number={1},
  pages={1--10},
  year={2022},
  publisher={Nature Publishing Group}
}