-
Notifications
You must be signed in to change notification settings - Fork 17
OWL Axiomatization
By encoding phenotypes in OWL we can leverage reasoners to automatically construct and verify ontologies, and we can use the axioms to connect data across different OBO library ontologies.
The motivation and background for the OWL axiomatization of phenotypes can be found in:
C J Mungall, Georgios Gkoutos, Cynthia Smith, Melissa Haendel, Suzanna Lewis, Michael Ashburner (2010) Integrating phenotype ontologies across multiple species _Genome Biology 11 (1)
The essence of the axiomatization is that a subset of phenotypes are assigned equivalence axioms to a class expression whose signature includes a [PATO] class (the Quality, or "Q") and one or more classes from OBO library ontologies (the Entity, or "E"), such as biological processes or anatomical entities.
The model we use was originally described in [Representing phenotypes in OWL http://ceur-ws.org/Vol-258/paper29.pdf]. We are currently using a slight variation of this model (the "subquality" model, such that all phenotypes are defined as
With variants on this pattern for more complex phenotypes that do not fit a simple EQ pattern.
This differs from the original pattern in that we enclose all descriptions in a "has_part" expression. This is to be consistent with phenotype ontologies in which the subclass hierarchy does not follow the PATO hierarchy
The modeling patterns we use can be found here and here.
A number of other promising models have been proposed, such as the 'phene' model, as described in:
Hoehndorf, Robert, Anika Oellrich, and Dietrich Rebholz-Schuhmann. Interoperability between phenotype and anatomy ontologies Bioinformatics 26.24 (2010): 3112-3118.
Our choice of model has been influenced by many practical factors, including ease of authoring using standard tools and speed of reasoning.
Historically we have authored the axioms in obo format, which has imposed severe practical limits of expressivity, and the existing simple pattern comes in part from this constraint. However, even as we move into directly authoring the axioms in OWL, we want to keep the axioms simple enough that a non-logician can understand and use them whilst still using standard OWL environments such as Protege
The resulting combined ontology set we reasoner over can be large, so it is important we use fast reasoners such as Elk. We therefore avoid using constructs such as OWL cardinality, and instead use PATO classes to define absence, duplication etc. It appears we do not lose anything by going this route.