ChEBI

ChEBI (Chemical Entities of Biological Interest) is two things in one: a naming dictionary of small molecules and an ontology that records how those molecules relate to each other. Each entry pins a compound to a stable identifier and gathers its synonyms, chemical structure, formula, and cross-references to other databases. The ontology layer adds relationships such as "is a", "has role", and "is conjugate base of", so you can walk from a specific molecule up to its broader class or across to its function.

You can download the whole thing from EMBL-EBI. Structures come as `SDF` files (a standard chemistry format that carries the molecular graph), and the ontology comes as `OWL`, which loads into any RDF or graph tool. There is also a web service and a SPARQL endpoint if you would rather query than bulk-load. For retrieval you rarely embed ChEBI as prose. The usual pattern is to treat it as a lookup table: normalise a chemical mention in your text to its ChEBI identifier, then attach that identifier as metadata on the chunk so filters and joins line up.

ChEBI is best for entity linking and query expansion in chemistry or biomedical RAG. If your documents mention "aspirin", "acetylsalicylic acid", and "2-acetoxybenzoic acid", ChEBI ties all three to one identifier so retrieval treats them as the same substance, and the ontology lets you broaden a query up to the related class.

Watch the scope: ChEBI deliberately focuses on small molecules of biological interest, so it is not a complete chemical catalogue, and being hand-curated it is modest in size next to automated collections. The licence is refreshingly clean, though. CC BY 4.0 means commercial use is fine as long as you credit ChEBI, with no share-alike strings attached.

Think of ChEBI as the controlled vocabulary and PubChem or ChEMBL as the bulk data. Pairing them works well: ChEMBL or PubChem for coverage and measured activity, ChEBI for consistent naming and structure.