Status: intermediate draft

Authors and contributors:

• Chris Mungall (author)

Date: 2012

Document Type: ontology_design_pattern

This article describes the practical approach taken in Uberon to avoid asserted multiple inheritance, and documents where exceptions are made to standard normalization practice

It is good practice to avoid asserted multiple inheritance in ontology development. Instead, classes should have a single asserted superclass or an equivalence axioms containing a single named class (the 'genus') plus one or more additional clauses (the 'differentiae').

In addition, if the full practice of normalization is being followed, then the asserted parent hierarchy should be jointly-exhaustive and pairwise disjoint.

If ontology develops follow this methodology, then their ontologies become easier to maintain

Uberon follows this methodology, but not dogmatically. Exceptions are made in various circumastances:

1. If the exact normalization strategy has not yet been decided, then we deem it acceptable to delay normalization and assert multiple parents.

2. Many of the axioms in this ontology are annotated with provenance information. The goal is to eventually have detailed evidence and provenance information for all axioms. In general it is preferable that this information is attached to non-subclass relationships. However, in cases where another ontology has a subclass axiom that can be generalized to Uberon, it is most straightforard to represent this as a subclass axiom with a provenance note pointing back to the source ontology. This can lead to multiple superclass assertions - in some cases, one will be redundant with another, but we want to retain both as there is separate evidence for both. In other cases, the parents will be non-disjoint classes with distinct evidence for both.

Over time, there will be less of a need for asserted multiple parentage, but the current strategy is only to eliminate multiple inheritance where it helps with ontology maintenance.

Note that the inferred hierarchy of this ontology is very lattice-like due to grouping classes and inferred multiple inheritance. This can be problematic for some visualization tools. Our strategy here is to provide means of extracting subsets of the ontology that exclude these grouping classes, but this is considered a separate issue to asserted multiple inheritance.

• http://odps.sourceforge.net/odp/html/Normalisation.html
• http://ontogenesis.knowledgeblog.org/49