GRAPPA - Graph Neural Network for Predicting the Parameters of the Antoine equation

GRAPPA is a machine learning model based on a graph neural network architecture that allows predicting the parameters of the Antoine equation only based on molecular structure. This repository contains the trained model and examples to show how to calculate vapor pressure, the underlying Antoine parameters and boiling temperatures. The model only requires the SMILES representation of the molecule as input.

GRAPPA is based on the Antoine equation in the form:

$$ \ln(p^\mathrm{s} / \mathrm{kPa}) = A - \frac{B}{T / \mathrm{K} + C} $$

where $p^\mathrm{s}$ is the vapor pressure, $T$ is the temperature, and $A$, $B$, and $C$ are the Antoine parameters. A detailed description of the model, its appicability ranges and the expected prediction accuracies can be found in the preprint available on arXiv.

Installing GRAPPA

1. Clone the repository

   git clone https://github.com/marco-hoffmann/GRAPPA.git
   cd GRAPPA

2. Create an environment with all required packages.

   To install the required packages, simply create a new conda environment from the grappa_env.yml file:

   conda env create -f grappa_env.yml

3. Activate the environment

   conda activate grappa_env

Using GRAPPA

The notebook file GRAPPA_examples.ipynb contains examples on how to use the GRAPPA model to predict vapor pressure, Antoine parameters and boiling temperatures.

Citing GRAPPA

If you use GRAPPA in your research, please cite the following publication:

@misc{Hoffmann2025,
      title={GRAPPA - A Hybrid Graph Neural Network for Predicting Pure Component Vapor Pressures}, 
      author={Marco Hoffmann and Hans Hasse and Fabian Jirasek},
      year={2025},
      eprint={2501.08729},
      archivePrefix={arXiv},
      primaryClass={cs.LG},
      url={https://arxiv.org/abs/2501.08729}, 
}

License

The project is licensed under the MIT License. See the LICENSE file for more information.