Predicting solubility curves in solvent mixtures using thermodynamic cycles and machine learning

This repository contains the code used in our work on predicting solubility of organic compounds in binary solvent mixtures using machine learning.

• SolProp_ML/ — The SolProp-mix software tool used in this work along with the weights. See the README inside that folder for installation instructions.
• DirectML/ — Python scripts for Chemprop and fastsolv, both extended to handle mixed-solvent systems, along with the training data and the weights of Chemprop model.

If you use SolProp-mix, please cite our work: Predicting solubility curves in solvent mixtures using thermodynamic cycles and machine learning Simona Buzzi, Emad Al Ibrahim, Ulderico Di Caprio, William H. Green, Florence Vermeire — ChemRxiv, 2026

@article{Buzzi2026solubility,
  author  = {Simona Buzzi and Emad Al Ibrahim and Ulderico Di Caprio and William H. Green and Florence Vermeire},
  title   = {Predicting solubility curves in solvent mixtures using thermodynamic cycles and machine learning},
  journal = {ChemRxiv},
  year    = {2026},
  doi     = {10.26434/chemrxiv.15003912/v2}
}

For questions or feedback, please contact:

• Simona Buzzi