Modeling paired structures Design Pattern

Representation of paired structures

Status: intermediate draft

Authors and contributors:

• Chris Mungall (author)

Date: 2012

Document Type: ontology_design_pattern

Abstract

...

Pattern

Classes for bilaterally paired structures should in general follow this subsumption hierarchy:

• X
  • left X
  • right X

Often it is not necessary to include the two subclasses in the ontology. See the guidelines section below

Laterality relations

When left and right subclasses are materialized, they should have logical definitions. The following relation hierarchy should be used for bilaterally paired structures:

• in_lateral_side_of
  • in_left_side_of
  • in_right_side_of

For example:

• left putamen EquivalentTo putamen and in_left_side_of some brain

However, these will not always be appropriate. For example, it may make sense to define certain paired veins by the structure they connect to. E.g. left testicular vein EquivalentTo 'testicular vein' and connected_to some left testicle.

For many paired structures in Bilateria, the class multi-cellular organism can be used

When to create subclasses

It is not always necessary to create the left and right subclasses. This is particularly true for skeletal subdivisions, which may more typically be symmetric. Internal organs may have more asymmetries and thus there is more of a case for making subclasses.

Sometimes it can be useful to include both subclasses for structures that meet in the midline, so as to describe the connecting structure. For example:

• mandibular symphysis connects some 'left dentary' and connects some 'right dentary`

Although there are other ways to say the equivalent thing, the above pattern is recommended in this ontology, as it is the most straightforward.

In general, if a one half is materialized in the ontology, the other should be - especially for symmetric structures

Generic parent class

The generic class may also receive a in_lateral_side_of axiom. For example, if every naris is part of either the left or right side of the head, then we can write an axiom:

• naris SubClassOf in_lateral_side_of some head

However, if there exist unpaired medial nares this would be inappropriate.

Errors can automatically be detected for example if there is a class 'median naris' and this is stated to be in the medial part of the head.

Pair classes

For any type of structure that appears bilaterally, it is possible to create 4 distinct classes:

• X
• left X
• right X
• pair of Xs

It is important not to conflate these. A pair of Xs is not the same thing as an X. Example: pair of lungs. The axiom pattern is:

• X SubClassOf part_of some pair_of_Xs
• {left,right} X SubClassOf X
• {left,right} X SubClassOf in_{left,right}_side_of some pair_of_Xs

In general the class 'pair_of_Xs' can be omitted from the ontology for symmetric structures, although these sometimes included, particularly in cases where other ontologies create the pairing class.

Inference

A standard EL reasoner will produce results that are always valid. For example, a left CA1 field of hippocampus will be inferred to be a subclass of left hippocampal field, assuming these are defined appropriately (ie using an in_lateral_side_of axiom).

However, inferences may not always be complete. If we have:

'left hippocampal field' = 'hippocampal field' and in_left_side_of some brain
'hippocampal field' SubClassOf part_of some 'Ammons horn'
'left Ammons horn' = 'Ammons horn' and in_left_side_of some brain

We do not get the inference:

'left hippocampal field' SubClassOf part_of some 'left Ammons horn'

Of course, this can be circumvented via:

'left hippocampal field' = 'hippocampal field' and in_left_side_of some 'Ammons horn'

But this introduces a new problem, namely that we are asserting the partonomy in two places -- if we need to change the partonomy at source we must manually change the partonomy in the compositional terms; this is precisely the kind of issue a reasoner is intended to circumvent.

This is a known issue with basic reasoning.

In this particular case, the issue could be circumvented by having an entirely automated template-based process for lateral structures. This would fix other issues of repetition, e.g. in synonyms.