BChemXtract: Automated Extraction and FAIRification of Chemical Structures from Scientific Publications

Chemical structures are published as images, which makes them inappropriate for database indexing, structure search and computational reuse. Recognition tools recover them after publication and cannot restore what the rendering discarded. BChemXtract is an open-source Java library that extracts structures directly from the ChemDraw CDX and CDXML files authors already submit. It separates chemical fragments from annotations, expands abbreviations from a curated lookup table and converts the structures with the Chemistry Development Kit into standard identifiers. In the Beilstein Journals the extracted structures are validated by hand, embedded in the article metadata as Schema.org JSON-LD and deposited in PubChem. Volume 20 of the Beilstein Journal of Organic Chemistry (BJOC), 2993 figures and schemes, yielded 23,320 structures with InChIs in under two minutes. On a 225-file benchmark 98.2% of unique substances got a valid InChI, against 14.2% with Open Babel. MIT licence, with a web application and the benchmark corpus. Library: github.com/beilstein-institut/bchemxtractWeb application: cheminfo.beilstein.org/bchemxtract