cmungall
commented
9 years ago Thanks for the report. The OWLAPI is the correct place to make any fix to the parser (the source in this repo is legacy). However, this is the correct tracker to use for clarifications in the spec, and a clarification is indeed required here.
The BNF allows for quoted strings after an xref, but there is no translation of this to OWL specified. The guide is more expansive, but doesn't really provide any semantics. It does note that this feature is discouraged.
There are 3 possibilities here:
- The quoted strings are treated like
!s and are silently discarded - The quoted strings are treated as rdfs:labels for the entities referenced by the dbxref
- The quoted strings are treated as axiom annotations on the dbxref annotation
Note that some ontologies like to retain their dbxref strings, and are reliant on the owlapi for roundtripping, so would be displeased with 1.
The assumption we implicitly made in the spec was that the strings would be used as labels, and thus 2 is valid. This indeed works for the two main ontologies reliant on this that I am aware of GO and HP.
However, CHEBI violates this unwritten assumption. Translation 3 would be safe here, as people could put whatever they like and the string would always be scoped to that particular assertion. This would work fine for GO and HP too.
So there is a strong argument for (3). However, there are some things to be aware of. Because a dbxref can appear as provenance in a def or synonym annotation, we have potentially depth=2 of axiom annotations. We would like to do tests on all OWL serializations and parsers on this since AFAICT this has never been used in the wild before.
Alternatively, if we adopt (2) as official then CHEBI will have to be changed.
Comments? @dosumis?
Either way, I will contact CHEBI and see if they are open to modifying their obo export and simply drop the strings. I don't think they serve any purpose to anyone, as they just repeat the ID space of the xref.
ignazio1977
leechuck
This is a copy of https://github.com/owlcs/owlapi/issues/415; I think the issue may be better resolved here and then merged into owlapi.
For some ontologies in OBO format, xref clauses in the OBO file result in rdfs:label annotations to the class, with a part of the xref as value.
Example code (using OWLAPI 4.1.0-RC2, but also present in earlier 4.X versions):
The desired output is:
Annotation(rdfs:label "paliperidone"^^xsd:string)The output I receive is:
The annotations are generated from the following statements in the OBO file:
i.e., the rdfs:label is created from the source, in quotes, after the xrefs.
The problem seems to be in line 1210 of Obo2Owl.java (https://github.com/owlcollab/oboformat/blob/gh-pages/src/main/java/org/obolibrary/obo2owl/Obo2Owl.java#L1210, or https://github.com/owlcs/owlapi/blob/b2401a615a1151244a48b5c0ce88a5a98995a678/oboformat/src/main/java/org/obolibrary/obo2owl/OWLAPIObo2Owl.java#L1392 in OWLAPI) where an rdfs:label is created based on xref values. This breaks identifying the label for several ontologies in OBO format.