HOME > DScribe

DScribe

DScribe is an open-source Python library for generating feature descriptors from atomic structures. It provides efficient implementations of commonly used representations in materials informatics and machine learning, such as the Coulomb matrix, SOAP (Smooth Overlap of Atomic Positions), ACSF (Atom-centered Symmetry Functions) and MBTR (Many-body Tensor Representation). By transforming atomic configurations into numerical vectors, DScribe enables the integration of machine learning models with quantum mechanical calculations, facilitating high-throughput screening and surrogate model construction.

Information
Case / Article
Learning Content
Review
Developer’s Voice

Information

Official site	https://singroup.github.io/dscribe/latest/<
Openness	★★★
License	Apache License 2.0
Core Developers	Developers: Aalto University, Department of Applied Physics by the Computational Electronic Structure Theory (CEST) and Surfaces and Interfaces at the Nanoscale (SIN) groups.
Related Papers	Marc Philipp Bahlke, Natnael Mogos, Jonny Proppe, Carmen Herrmann, “Exchange spin coupling from gaussian process regression,” The Journal of Physical Chemistry A, 2020. doi:10.1021/acs.jpca.0c05983 Annika Stuke, Milica Todorović, Matthias Rupp, Christian Kunkel, Kunal Ghosh, Lauri Himanen, Patrick Rinke, “Chemical diversity in molecular orbital energy predictions with kernel ridge regression,” The Journal of Chemical Physics, 150(20):204121, 2019. doi:10.1063/1.5086105
Other	Description: DScribe is widely adopted in computational materials science for constructing machine learning potentials and property prediction models. Its modular API allows seamless integration with major Python frameworks such as scikit-learn, PyTorch, and TensorFlow. DScribe has become an essential component of modern materials informatics pipelines, enabling the efficient generation of descriptors for tasks ranging from structure–property relationships to active learning in molecular and crystal simulations.