robot convert from OWL to OBO drops AnnotationProperties and NamedIndividuals...

[edwardsnj]

NOTE: repost of ticket robot ticket #1276 per suggestion of @matentzn

• ROBOT version 1.9.8 (docker-based)
• Command: robot convert --input "GNOme.owl" --format obo --output "GNOme.obo"

I get term stanzas in the OBO file like this:

[Term]
id: GNO:10000273
name: glycan of molecular weight 164.07 Da
def: "A glycan characterized by underivitized molecular weight of 164.07 Daltons." []
is_a: GNO:00000001 ! glycan
property_value: GNO:00000021 GNO:00000012
property_value: GNO:00000024 GNO:00000001

Which is correct, but the AnnotationProperties (GNO:00000021 and GNO:00000024) are not output as their own stanzas (with id, name, def, etc.) and neither are NamedIndividuals (GNO:00000012, ...).

Perhaps this is as intended due to OBO format restrictions, but I was surprised they were missing.

Thanks!

[cmungall]

Individuals are not supported but APs are, can you provide the input?

[edwardsnj]

Any of the GNOme OWL files will demonstrate the issue.

Current release

[matentzn]

Minimal example:

<?xml version="1.0" encoding="UTF-8"?>
<rdf:RDF
    xmlns:rdf="http://www.w3.org/1999/02/22-rdf-syntax-ns#"
    xmlns:rdfs="http://www.w3.org/2000/01/rdf-schema#"
    xmlns:owl="http://www.w3.org/2002/07/owl#"
    xmlns:obo="http://purl.obolibrary.org/obo/"
    xmlns:oboInOwl="http://www.geneontology.org/formats/oboInOwl#">

  <!-- Annotation property declarations -->
  <owl:AnnotationProperty rdf:about="http://purl.obolibrary.org/obo/GNO_00000021">
    <rdfs:label>has subsumption level</rdfs:label>
  </owl:AnnotationProperty>

  <owl:AnnotationProperty rdf:about="http://purl.obolibrary.org/obo/GNO_00000024">
    <rdfs:label>has base composition type</rdfs:label>
  </owl:AnnotationProperty>

  <!-- GNO:10000273 as an OWL Class -->
  <owl:Class rdf:about="http://purl.obolibrary.org/obo/GNO_10000273">
    <rdfs:label>glycan of molecular weight 164.07 Da</rdfs:label>
    <obo:IAO_0000115>A glycan characterized by underivitized molecular weight of 164.07 Daltons.</obo:IAO_0000115>
    <rdfs:subClassOf rdf:resource="http://purl.obolibrary.org/obo/GNO_00000001"/>
    <obo:GNO_00000021 rdf:resource="http://purl.obolibrary.org/obo/GNO_00000012"/>
    <obo:GNO_00000024 rdf:resource="http://purl.obolibrary.org/obo/GNO_00000001"/>
  </owl:Class>

  <!-- GNO:00000001 as both Class and NamedIndividual (OWL 2 punning) -->
  <owl:Class rdf:about="http://purl.obolibrary.org/obo/GNO_00000001">
    <rdfs:label>glycan</rdfs:label>
  </owl:Class>

  <!-- Named individuals -->
  <owl:NamedIndividual rdf:about="http://purl.obolibrary.org/obo/GNO_00000001">
    <rdfs:label>glycan</rdfs:label>
  </owl:NamedIndividual>

  <owl:NamedIndividual rdf:about="http://purl.obolibrary.org/obo/GNO_00000012">
    <rdfs:label>molecular weight level</rdfs:label>
  </owl:NamedIndividual>

</rdf:RDF>

Command (ROBOT 1.10):

robot convert -i obo-gno-test.owl -o obo-gno-test.obo

Output:

format-version: 1.2
idspace: oboInOwl http://www.geneontology.org/formats/oboInOwl# 
ontology: 

[Term]
id: GNO:00000001
name: glycan

[Term]
id: GNO:10000273
name: glycan of molecular weight 164.07 Da
def: "A glycan characterized by underivitized molecular weight of 164.07 Daltons." []
is_a: GNO:00000001 ! glycan
property_value: GNO:00000021 GNO:00000012
property_value: GNO:00000024 GNO:00000001

The key is that property_value: GNO:00000024 GNO:00000001 should work. The other one we could argue is out of scope for OBO format because OBO explicitly cannot handle named individuals.

[gouttegd]

The OWL2OBO translator in the OWLAPI will produce a Typedef frame for an annotation property only if the property is explicitly annotated with a oboInOwl:is_metadata_tag.

<owl:AnnotationProperty rdf:about="http://purl.obolibrary.org/obo/GNO_00000021">
    <rdfs:label>has subsumption level</rdfs:label>
    <oboInOwl:is_metadata_tag rdf:datatype="http://www.w3.org/2001/XMLSchema#boolean">true</oboInOwl:is_metadata_tag>
</owl:AnnotationProperty>

will produce the expected frame:

[Typedef]
id: GNO:00000021
name: has subsumption level
is_metadata_tag: true

Why this behaviour? Why requiring that an annotation property must be annotated with oboInOwl:is_metadata_tag in order to be translated, instead of simply translating any annotation property into a Typedef frame?

I don’t know for sure, but my guess is that this was done this way for two reasons:

(A) Because we may not actually want to translate any annotation property into a OBO Typedef frame. For example, we don’t necessarily want to end up with a Typedef frame for rdfs:label.

(B) Because upon translating an annotation property into a Typedef frame, the OWL2OBO translator would need to inject itself a oboInOwl:is_metadata_tag annotation on the property if such an annotation is not already present. This would be required because both object properties and annotation properties are translated into the same type of OBO frame (Typedef), and only the is_metadata_tag allows to distinguish between the two types of property. Not injecting the oboInOwl:is_metadata_tag would lead the resulting Typedef frame to be later interpreted as a frame describing an object property. But this would still be a form of axiom injection, where the OWL2OBO translator would fabricate an axiom that does not actually exist in the original ontololy. I can understand that whoever wrote that part of the OWL2OBO translator felt it was better to drop an annotation property than to inject an annotation assertion axiom to it – that seems like the correct decision to me.

Bottom line is that, unless we agree to revise that behaviour, whoever authors an ontology in a non-OBO format but wants to export it to OBO should take care of explicitly marking their own annotation properties with oboInOwl:is_metadata_tag.

[edwardsnj]

OK, I implemented this at the end of last week. In the resulting OBO format, I have stanzas for Typedefs corresponding to the AnnotationProperties, which is great. In the Class stanzas, some of the properties are now presented as follows:

property_value: GNO:00000024 GNO:G32390NS

becomes

relationship: GNO:00000035 GNO:G32390NS ! has_topology G32390NS

which is great. However, those that are not URLs or resolvable URI references, such as strings:

property_value: GNO:00000102 "164" xsd:string

are unchanged. I'm not sure why these do not get the same treatment as the relationships...

property_value: GNO:00000102 "164" xsd:string ! has_structure_characterization_score 164

Nevertheless, the situation for generating OBO files from OWL is very much improved. Any suggestions for how to get something into the OBO file to represent the NamedIndividuals? I don't like that they are dangling references to the ontology itself that cannot be resolved without external information.

Cheers!