Status: intermediate draft

Authors and contributors:

• Chris Mungall (author)

Date: 2012

Document Type: ontology_design_pattern

...

OWL reasoners can automatically generate a subsumption hierarchy. However, we often want to automatically classify a partonomy - for example:

• manual digit phalanx SubClassOf part_of some forelimb skeleton

Our solution here is to create a bone grouping class for every skeletal subdivision. For example:

1. manual phalanx EquivalentTo phalanx and part_of some manus
2. phalanx SubClassOf some bone
3. forelimb bone EquivalentTo bone and part_of some pectoral limb
4. forelimb bone SubClassOf part_of some forelimb skeleton

With the above axioms, manual phalanx will be classified as forelimb bone, which is asserted to be part of the forelimb skeleton.

The combination of 3+4 above is an example of a hidden GCI. We create these for every skeletal subdivsion, such that we have:

• X bone EquivalentTo bone and part_of some X
• X bone SubClassOf part_of some X skeleton

It is important that X is the organism subdivision here.

The generalized solution is to create an element grouping class (e.g. bone, or more generally, skeletal element, or a specific element type, such as fin radial or autopod phalanx) for every skeletal subdivision.

1. LOCATION ELEMENT EquivalentTo ELEMENT and part_of some LOCATION
2. ELEMENT SubClassOf some skeletal element
3. LOCATION skeletal element EquivalentTo skeletal element and part_of some LOCATION
4. LOCATION skeletal element SubClassOf part_of some LOCATION

One alternative is to use an actual GCI. Here we 'unfold' axioms 3 and 4 above to make:

1. manual phalanx EquivalentTo phalanx and part_of some manus
2. phalanx SubClassOf some bone
3. bone and part_of some pectoral limb SubClassOf part_of some forelimb skeleton

This is equivalent, but there are practical strengths of benefits to both approaches:

• hidden GCIs of the form above are possible in obo-format; actual GCIs are not
• many ontologies like MA have 'X bone' classes, whereas ontologies like FMA have skeletal subdivsions
• materializing both forms in Uberon simplifies bridging between these
• the proliferation of terms can be confusing to annotators, but it is possible to make subsets that exclude either form