A Dataset for Relation Extraction of Natural-Products

A curated evaluation dataset for end-to-end Relation Extraction of relationships between organisms and natural-products.

Details about the manual annotation:

• For Chemicals:

  • The chemical labels are annotated as they appear in the abstract.

  • In abstracts, singular chemicals and classes of chemicals produced by a specific organism were distinguished.

  • The "type" attribute {“chemical”, “class”} is used to indicate the nature of the mentioned name.

  • A "class" attribute for chemical entities has also been included if class information is present in the abstract.

  • A Wikidata and PubChem identifiers were assigned to chemicals and classes when available.

• For Organisms:

  • The organism labels are annotated as they appear in the abstract.

  • If in an abstract, the genus name was mention first, e.g. "Plakinastrella sp." and then the specie name e.g "Plakinastrella clathrata" is precise, then only the specie name is used.

  • A Wikidata identifier was assigned to all organisms.

  • In some abstracts, only the genus name is mentioned.

• For Relations:

  • Only the relations explicitly mentioned in the abstract are reported in the output labels.

  • Relations are reported in their order of appearance in the abstract.