Active Learning of Molecular Data for Task-Specific Objectives

Python Requirements

latest versions of the following packages:

• NumPy
• Pandas
• Matplotlib
• SciPy
• SKLearn

Downloading the data

Before running the script, ensure you download the relevant dataset and save it in a directory titled data with one subdirectory for each dataset. The datasets used in the publication can be found in the links below:

• AA : https://doi.org/10.5281/zenodo.3967308.
• OE62 : https://doi.org/10.5281/zenodo.4035923.
• QM9 : https://doi.org/10.5281/zenodo.4035918.

Instructions

To run the code, you need to specify the configuration in a config file (c.f. example). Remember to modify the mbtr_path and json_path to the appropriate paths with the MBTR vectors and HOMO energy values of the training dataset.

Invoking __main__.py (link) with the config file will execute the specified active learning setup. All the output data and log files are saved in the working directory.

To run multiple runs of an active learning loop, please use the cookie-cutter template defined here. The template will generate the folder structure and relevant config files.

Cite Us

https://doi.org/10.1063/5.0229834

@article{ghosh_etal_2025,
    author = {Ghosh, Kunal and Todorović, Milica and Vehtari, Aki and Rinke, Patrick},
    title = {Active learning of molecular data for task-specific objectives},
    journal = {The Journal of Chemical Physics},
    volume = {162},
    number = {1},
    pages = {014103},
    year = {2025},
    month = {01},
    issn = {0021-9606},
    doi = {10.1063/5.0229834},
    url = {https://doi.org/10.1063/5.0229834},
    eprint = {https://pubs.aip.org/aip/jcp/article-pdf/doi/10.1063/5.0229834/20327578/014103\_1\_5.0229834.pdf},
}