HANNA

HANNA is a machine learning model for predicting the excess Gibbs energy of the liquid phase in mixtures with an arbitrary number of components. Through automatic differentiation, HANNA derives thermodynamically consistent activity coefficients from the excess Gibbs energy. HANNA is trained to more than 800,000 experimental data points of vapor-liquid equilibria, liquid-liquid equilibria, activity coefficients at infinite dilution and excess enthalpies in binary mixtures. As input, only the SMILES notation of all molecules and the considered state point (composition and temperature) is required. In comprehensive benchmarks, HANNA was found to yield better results than the state-of-the-art models of the UNIFAC family and other ML models.

Note: This repository is based on the HANNA prototype implementation by tspecht93, which was restricted to binary mixtures and not trained on liquid-liquid equilibrium data or excess enthalpies; the corresponding original paper is available here.

Easy Use

You can explore HANNA and other models from our working group on our interactive web interface, MLPROP, without any installation.

Installing and using HANNA

To set up the project, follow these steps:

1. Clone the repository:

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

2. Create the conda environment:

   Use the provided .yml file to create the conda environment.

   conda env create -f environment.yml

   Note: If you want to run the model on a GPU, make sure to install the PyTorch version with CUDA support.

3. Test the installation:

   Open the demo.ipynb notebook and run the cells using the HANNA environment to test the installation.

   If you obtain output plots similar to the following, the installation was successful.

   Binary Prediction:

   

   Ternary Prediction:

   

License

This project is licensed under the MIT License. See the LICENSE file for details.

Tested Versions

The software has been tested with the following package versions:

• python==3.10.19
• pytorch==2.10.0
• numpy==2.2.6
• pandas==2.3.3
• rdkit==2025.9.5
• transformers==5.1.0
• tokenizers==0.22.2
• ipykernel==7.2.0
• ipywidgets==8.1.8
• scikit-learn==1.7.2
• matplotlib==3.10.8