Barry Smith’s comments on “Semi-automated ontology generation within OBO- Edit” paper

A new paper entitled "Semi-automated ontology generation within OBO-Edit" has just appeared in the journal Bioinformatics [1]. The paper describes a system called 'DOG4DAG', which supports the creation and extension of OBO ontologies by semi- automatically generating terms, definitions and parent-child relations from text in PubMed, the web and PDF repositories. I can well imagine that the DOG4DAG system may have some value in the creation of extensions to ontologies which already exist, and which have been created through a careful process of manual editing. When it comes to ontology generation, however, I am not optimistic that the results would be of high value. Automatic ontology generation, thus far, has a poor record, and the illustrations of small fragments of generated ontologies provided in the paper do not inspire confidence that we are here dealing with something essentially new.

Even were DOG4DAG to produce good results, however, I would still be sceptical as to its value as a tool for ontology generation. Ontology works if it helps to break down walls between silos. This means that there should be rather few, well-constructed and well-tested and thoroughly criticized, ontologies, in order to ensure that, for a given body of data, the same ontologies are used, and progressively enhanced, by all of these working in the relevant field. To enable too easy generation of new ontologies at the drop of a hit

would risk defeating the very purpose of ontologies. It would be rather like allowing new versions of English, incommensurable with the version we are all currently trying to speak, to be created by new groups of researchers. Or rather like allowing new systems of measurement units to be created in ad hoc ways by different groups of scientists on the fly. This would not promote good science.


[1] Wächter T, Schroeder M. Semi-automated ontology generation within OBO-Edit. Bioinformatics, 2010 Jun 15;26(12):88-96.