Automated Extraction and FAIRification of Chemical Structures from Scientific Publications
- 1Beilstein-Institut zur Förderung der Chemischen Wissenschaften, Trakehner Straße 7–9, Frankfurt am Main, Germany, 60487
- 2Natural Product Cheminformatics, Friedrich Schiller University Jena, Rosalind-Franklin-Straße 5 , Jena, Germany, 07745
Abstract
Chemical structures are published predominantly as images, which makes them inaccessible for database indexing, structure-based search and computational reuse. While Optical Chemical Structure Recognition and PDF-extraction tools attempt post-publication recovery, they cannot restore the information lost during image rendering. To address this, we present BChemXtract, an open-source Java library that extracts chemical structures directly from the CDX and CDXML files that authors already submit with their manuscripts. BChemXtract parses the CDX object hierarchy to isolate chemical fragments from graphical annotations and expands abbreviations using a community-curated lookup table. The structures are then converted via the Chemistry Development Kit into standard machine-readable formats. BChemXtract is integrated into the production workflow of the Beilstein Journals, where extracted structures are manually validated before being embedded in article metadata as Schema.org JSON-LD, deposited in PubChem and offered for download. Applied to the 2993 ChemDraw figures and schemes of volume 20 of the Beilstein Journal of Organic Chemistry, it yielded 23,320 structures with InChIs in under two minutes. On a 225-file benchmark corpus, it generated valid InChIs for 98.2% of extracted unique substances, compared with 14.2% for Open Babel, at comparable per-file speed. BChemXtract makes published chemistry FAIR at the point of publication without altering author workflows. It is available under the MIT license together with a web application and the benchmark corpus.
Related
- Toward Autonomous Extraction and Building of Machine-Readable Molecular Models from Publications Using Large Language Models
- C-MAGE: Automated Extraction and Organization Engine for Molecular Representations in Chemical Documents
- Benchmarking thermoelectric property extraction from scientific text
- Resolving Positional Markush Structures for Chemical Reaction Extraction
- Chemical Structures Cleaning Center: An Open Web Service for Automated Chemical Databases Curation
BibTeX
@misc{bansch2026automated,
title = {{Automated Extraction and FAIRification of Chemical Structures from Scientific Publications}},
author = {Felix Bänsch and Kohulan Rajan and Markus Nietfeld},
howpublished = {ChemRxiv},
year = {2026},
month = oct,
doi = {10.26434/chemrxiv.15010166/v1},
url = {https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010166/v1},
note = {CC BY},
}
RIS
TY - UNPB
TI - Automated Extraction and FAIRification of Chemical Structures from Scientific Publications
AU - Felix Bänsch
AU - Kohulan Rajan
AU - Markus Nietfeld
PY - 2026
DA - 2026/10/09
PB - ChemRxiv
DO - 10.26434/chemrxiv.15010166/v1
UR - https://chemrxiv.org/doi/full/10.26434/chemrxiv.15010166/v1
AB - Chemical structures are published predominantly as images, which makes them inaccessible for database indexing, structure-based search and computational reuse. While Optical Chemical Structure Recognition and PDF-extraction tools attempt post-publication recovery, they cannot restore the information lost during image rendering. To address this, we present BChemXtract, an open-source Java library that extracts chemical structures directly from the CDX and CDXML files that authors already submit with their manuscripts. BChemXtract parses the CDX object hierarchy to isolate chemical fragments from graphical annotations and expands abbreviations using a community-curated lookup table. The structures are then converted via the Chemistry Development Kit into standard machine-readable formats. BChemXtract is integrated into the production workflow of the Beilstein Journals, where extracted structures are manually validated before being embedded in article metadata as Schema.org JSON-LD, deposited in PubChem and offered for download. Applied to the 2993 ChemDraw figures and schemes of volume 20 of the Beilstein Journal of Organic Chemistry, it yielded 23,320 structures with InChIs in under two minutes. On a 225-file benchmark corpus, it generated valid InChIs for 98.2% of extracted unique substances, compared with 14.2% for Open Babel, at comparable per-file speed. BChemXtract makes published chemistry FAIR at the point of publication without altering author workflows. It is available under the MIT license together with a web application and the benchmark corpus.
ER -