Evolutionary variability GCIs

Chris Mungall edited this page Jul 19, 2014 · 10 revisions

Evolutionary variability and general class axioms

Status: initial draft

Authors and contributors:

  • Chris Mungall (author)

Date: 2012

Document Type: ontology_design_pattern

Abstract

Multi-species ontologies may be axiomatically weaker than dedicated species ontologies, as all statements in an ontology must be universal. Here we describe how axioms can be scoped to a particular species or taxon using OWL general axioms

This is also described in the section 'Managing taxonomic variation' in the Uberon paper.

Problem

Frequently we are confronted with situations where we want to make a statement such as 'A REL some B', but this statement doesn't universally hold for all instances of A, as A encompasses structures present in a variety of taxa that vary in properties such as developmental lineage.

One possibility is to make subclasses of A and attach the axiom to those subclasses. For example, we can make subclasses of a generic eye, such as camera-type eye' orcompound eye`, and attach relationships to these. However, this approach can lead to ontology inflation where the subclasses are trivially distinguished.

Solution

Our solution here is to use general class axioms (aka General Class Inclusion axioms, or GCIs). This is logically equivalent to creating the subclass, but we use an OWL class expression in place of a named class:

   (A and part_of some TAXON-CLASS) SubClassOf REL some B

So if we want to make a statement that the adenohypophysis develops from Rathke's pouch in tetrapods, we would say:

  (adenohypophysis and part_of some Tetrapoda) SubClassOf develops_from some 'Rathkes pouch'

Note that we do not need to taxonomically scope the target class.

Examplar Class

See thymus primordium as an example (note: GCIs may not be visible in OntoBee). For best results view in Protege version 5:

Protege screenshot of thymus primordium

(the GCI viewer in Protege 4 is not great, you have to look at the 'about' tab)

This contains all the taxon-specific developmental relationships from table 13.1 of Kardong

OBO-Format implementation

OBO format has a very limited form of GCI support, but this is sufficient for taxon GCIs. In obo-format we would write:

   id: A
   relationship: REL B {gci_relation=part_of, gci_filler=NCBITaxon:nnnn}

This is provided as an authoring convenience. These relationships should be removed from obo-basic versions of the ontology, as they can possibly confuse some obo tools.

Existing taxon GCIs in Uberon

Currently these axioms of this form in Uberon come from two sources:

  1. Manually curated, often tagged with evidence from journal articles, textbooks and reviews
  2. Automatically generated, using axioms from species-specific ontologies

The former is available in standard editions of the ontology (with the exception of the 'basic' edition). The latter is only present in the 'composite' editions, such as 'composize-metazoan'. Here when we merge classes from a species-specific ontology (say ZFA, which is zebrafish specific) it isn't safe to merge in all of their relations, as they may not hold more generically. A classic case is the relationships to stages indicating when each structure arises (and possibly when it disappears) - these are highly specific. In these scenarios, safe taxon GCIs are added to the merge ontology.

For more details, see: Multi-species-composite-ontologies

See Also: