-
Notifications
You must be signed in to change notification settings - Fork 17
Modeling of Absence
- summary Discussion of issues pertaining to modeling of absence in phenotype ontologies
Much has been written on the subject of representing absence. Before diving into the logical issues it is worth examining patterns in existing phenotype ontologies to understand what user expectations may typically be for absence.
- Absence_Phenotypes_in_OWL (Phenoscape Wiki)
- (outdated) material on the old [PATO wiki](http://obofoundry.org/wiki/index.php/PATO:Revised_2008#Absence_and_counting
It is not uncommon to see patterns such as
From a strict logical perspective, this is inverted. "absent incisors" surely means "absence of all incisors", or put another way "the animal has no incisors". Yet it would be possible to have an animal with absent lower incisors and present upper incisors, yielding what seems a contradiction (because the subClass axiom would say this partial-incisor animal lacked all incisors).
If the ontology were in fact truly modeling "absence of all S" then it would lead to a curious ontology structure, with the typical tree structure of the anatomy ontology representing S inverted into a polyhierarchical fan in the absent-S ontology.
From this it can be cautiously inferred that the intent of the phenotype ontology curator and user is in fact to model "absence of some S" rather than "absence of all S". This is not necessarily a universal rule, and the intent may vary depending on whether we are talking about a serially repeated structure or one that typically occurs in isolation. The intent may also be to communicate that a significant number of S is missing.
It is also not uncommon to see patterns such as:
Again, from a strict logical perspective this is false. If the spleen is absent then what does the "morphology" of the parent refer to?
However, this inference is clearly a desirable one from the point of view of the phenotype ontology editors and users, as it is common in ontologies for a variety of structures. For example:
And:
These patterns can be formally defended on developmental biology grounds. "absence" here is not equivalent to logical absence. It refers specifically to developmental absence.
Furthermore, strict logical absence leads to undesirable inferences. It would be odd to include a nematode worm as having the phenotype "spleen absent", because worms have not evolved spleens. But the logical description of not having a spleen as part fets a worm.
Similarly, if the strict cardinality interpretation were intended, we would expect to see:
i.e. if you're missing your entire hindlegs, you're necessarily missing your femurs. But it must be emphatisized that this is not how phenotype ontologies are classified. This goes for a wide range of structures and other relationship types. In MP, "absent limb buds" are not classified under "absent limbs", even though it is impossible for a mammal to have limbs without having had limb buds.
The existing treatment of absence can be formally defended morphologically by conceiving of a morphological value space, with "large" at one end and "small" at the other. As we get continuously smaller, there may come an arbitrary point whereby we say "surely this is no longer a limb" (and of course, we are not talking about a pure geometrical size transformation here - as a limb reaches extreme edges of a size range various other morphological changes necessarily happen). But this cutoff is arguably arbitrary, and the resulting discontinuity causes problems. It is simpler to treat absence as being one end of a size scale.
This is barely touching the subject, and is intended to illustrate that things may be more subtle than naively treating words like "absent" as precisely equivalent to cardinality=0. An understanding of the medical, developmental and evolutionary contexts are absolutely required, together with an understanding of the entailments of different logical formulations.
Even though existing phenotype ontologies may not be conceived of formally, it is implicit than they do not model absence as being equivalent to cardinality=0 / not(has_part), because the structure of these ontologies would look radically different.
Link to Jim Balhoff's PhenoDay paper and discussion
Here's the link: http://phenoday2014.bio-lark.org/pdf/11.pdf