Bonding Analysis Database and Machine Learning Framework

Understanding the properties of a material often requires insight into its chemical bonding. Although electronic-structure calculations based on plane-wave density functional theory provide an accurate quantum-mechanical description of solids, they do not directly provide a chemically intuitive picture of bonding interactions. The LOBSTER program addresses this limitation by projecting plane-wave wavefunctions onto localized atomic orbitals, enabling the calculation of bond-resolved quantities that characterize interatomic interactions.

Performing such analyses has traditionally required multiple manual steps and significant expert intervention. To streamline this process, an automated workflow was developed and implemented in the atomate and atomate2 workflow libraries. The workflow combines density-functional-theory calculations with LOBSTER analyses, performs automated interpretation using the LobsterPy package, and stores results and associated metadata in a structured schema. Using this infrastructure, bonding analyses have been generated for approximately 13,000 materials and made openly available, providing a resource for chemical interpretation, materials discovery, and data-driven machine-learning studies.

The development of this workflow was motivated by the growing need to perform quantum-chemical bonding analyses at scales that are impractical using traditional manual approaches. While the LOBSTER program provides detailed insights into chemical bonding from plane-wave density-functional-theory calculations, its application has historically required substantial methodological expertise, multiple interconnected computational steps, and extensive manual post-processing. The workflow was developed to reduce these bottlenecks, standardize the interpretation of bonding data, and enable reliable high-throughput studies. Although initially motivated by challenges in automated materials discovery and data-driven materials science, it was conceived from the outset as a community resource rather than solely for internal use.

This work represents a collaboration between the Digital Materials Chemistry division at BAM, co-affiliated with FSU Jena, the BAM eScience group, collaborators at UCLouvain (Belgium), and the RWTH Aachen group responsible for the development of LOBSTER. The consortium combines expertise in quantum-chemical bonding analysis, electronic-structure theory, computational materials science, scientific software development, data science, machine learning, and research computing infrastructure.

The implementation builds on an ecosystem of open-source materials-science software, including pymatgen, atomate, atomate2, FireWorks, jobflow, jobflow-remote, and LobsterPy, to automate the generation, execution, analysis, and storage of bonding calculations. The workflow is designed for users familiar with electronic-structure calculations and supports large-scale studies through automated defaults, standardized analyses, machine-readable outputs, and documented workflows. High-performance computing resources were essential for generating the large-scale datasets underpinning the bonding database and its subsequent extension to approximately 13,000 materials.

Consistent with principles of open and reproducible science, the workflow, database infrastructure, and analysis software are publicly available under their respective licences. Validation relies on established LOBSTER quality metrics, projected density-of-states comparisons, oxidation-state assignments derived from bond-valence-based estimation methods, and coordination-environment analyses, while the associated software has undergone peer review through a dedicated research-software journal.

The workflow enabled large-scale quantum-chemical bonding analyses that would otherwise require substantial manual effort and specialist expertise, making systematic investigations of chemical bonding feasible across thousands of materials. By integrating automated calculation setup, bonding analysis, validation, and data management, it transformed a traditionally low-throughput methodology into a reproducible workflow suitable for data-driven materials science. The resulting bonding database grew from an initial collection of 1,520 compounds to approximately 13,000 materials, providing one of the largest openly available datasets of quantum-chemical bonding information for crystalline solids.

The infrastructure has supported a growing body of scientific work. The original workflow publication established automated bonding analysis as a high-throughput approach, the subsequent database publication demonstrated the generation and validation of large-scale bonding datasets, and the LobsterPy software publication provided automated interpretation, visualization, and descriptor-generation capabilities. More recently, the expanded database enabled a systematic assessment of bonding descriptors for machine-learning applications, demonstrating statistically significant relationships between bonding descriptors and vibrational, elastic, and thermal-transport properties.

Beyond method development, the workflow has been applied in collaborative studies of thermal and ionic transport in complex materials, where bonding analyses helped relate weak bonding, bonding heterogeneity, and anharmonicity to transport properties in Cu- and Ag-based argyrodites.

Openness and reproducibility are central design features. The workflow, database infrastructure, software, models, and analysis scripts are publicly available through open repositories, persistent identifiers, and documented data schemas, enabling complete reproduction of the workflow from data generation to descriptor extraction and model development. Uptake is reflected in a growing publication ecosystem that currently includes at least four dedicated peer-reviewed papers on workflow automation, database generation, software development, and descriptor assessment, alongside further downstream methodological and application-focused studies.

ML Application:  Aakash Ashok Naik, Nidal Dhamrait, Katharina Ueltzen, Christina Ertural, Philipp Benner, Gian-Marco Rignanese, Janine George, A Critical Assessment of Bonding Descriptors for Predicting Materials Properties, Advanced Intelligent Discovery 2026, 0, e70129. https://doi.org/10.1002/aidi.70129
Database: Naik, A.A., Ertural, C., Dhamrait, N. et al. A Quantum-Chemical Bonding Database for Solid-State Materials. Sci Data 10, 610 (2023). https://doi.org/10.1038/s41597-023-02477-5Workflows: J. George, G. Petretto, A. Naik, M. Esters, A. J. Jackson, R. Nelson, R. Dronskowski, G.-M. Rignanese, G. Hautier, ChemPlusChem 2022, 87, e202200123. https://doi.org/10.1002/cplu.202200123

Naik et al., (2024). LobsterPy: A package to automatically analyze LOBSTER runs. Journal of Open Source Software, 9(94), 6286, https://doi.org/10.21105/joss.06286

Ganose, A.M., Sahasrabuddhe, H., Asta, M., Beck, K., Biswas, T., Bonkowski, A., Bustamante, J., Chen, X., Chiang, Y., Chrzan, D.C. and Clary, J., 2025. Atomate2: Modular workflows for materials science. Digital discovery, 4(7), pp.1944-1973. https://pubs.rsc.org/dd/article/4/7/1944/888984/Atomate2-modular-workflows-for-materials-science

Ghata, A., Bernges, T., Maus, O., Wankmiller, B., Naik, A.A., Bustamante, J., Gaultois, M.W., Delaire, O., Hansen, M.R., George, J. and Zeier, W.G., 2024. Exploring the thermal and ionic transport of Cu+ conducting argyrodite Cu7PSe6. Advanced Energy Materials, 14(46), p.2402039. https://doi.org/10.1002/aenm.202402039