From NCBO Wiki
(Redirected from Mv:Shape)
Jump to: navigation, search

Currently the PATO shape hierarchy is rather flat, and contains many unusual terms and terms that lack strong definitions (many defs are sourced from dictionaries or wikipedia). This means that two curators, faced with the same phenotype, may select different terms: for example:

  • PATO:0001206 serrate Notched like a saw with teeth pointing toward the apex.
  • PATO:0001368 erose Having an irregularly toothed edge.

When we come to compare phenotypes algorithmically, erose and serrate are neither more similar than, say limaciform and tripartite

Even if curators tend to think exactly alike, and apply terms consistently, we still need a way of knowing that "tripartite" "T-shaped" and "Y-shaped" are extremely similar, and in fact a biological entity bearing any of these could easily transform it's shape into any of the others.



Full disclosure: many of these terms were placed in the hierarchy by me!

The current version of PATO has many problematic shape terms - this has been necessary in the first requirement-gathering phase; now it is time to put these terms in order. As an example of some of the problems:

Some defs use terms that are not themselves defined:

  • PATO:0001797 clubbed [DEF: "A shape quality inhering in a bearer by virtue of being plantarflexed, inverted and adducted."]

Some are circular

  • PATO:0001409 spindle-shaped [DEF: "A shape quality of having the shape of a spindle."]

Some are highly subjective, or definitions by simile

  • PATO:0001882 limaciform [DEF: "Resembling a slug."]

See discussion on botyoidal

Requirements for shapes in PATO

(feel free to add to this list)

Lack of ambiguity

When performing annotation, curators find it unsatisfying to feel they are forced into making arbitrary choices. Definitions should be clear and precise, and terms should be clearly delineated from one another. If the curator is describing a phenotype using a source that is itself vague, then there should be an appropriately general term or terms for the curator to use - they should not have to pick one arbitrily. This is the principle for all annotation: annotate to the level of precision in the hierarchy at which you can be confident, which leads us to the next requirement:

Hierarchical structure

is_a hierarchy

The current shape hierarchy is extremely flat. This makes it hard to clearly delineate terms from one another. Hierarchy is one of the main advantages of an ontology over a dictionary.

It is clear we need more hierarchy here - it is not clear what form this hierarchy should take

Inter-shape links

In addition to the is_a hierarchy, there may also be a need for links of some other type


There are simple shapes and complex shapes. The latter can be constructed from the former. This should be reflected in PATO somehow.

Ease of finding terms

Curators should not have to sift through a massive list of obscure terms checking their definitions.

This is more of a RQ for curators rather than end-users(?)


We want to use ontologies to help analyze data and find commonalities that would not otherwise be obvious. Thus there must be some kind of representation in PATO that allows similar shapes to be clustered together and dissimilar shapes to be distinguished. This also ties in with the following RQ

Reasoning and inference

In the simplest form this is subsumption reasoning: .e.g. queries for "tripartite" return annotations to "T-shaped".

It could also include more sophisticated kinds of inference: e.g.

Background Material

Collected PATO tracker items

This has been discussed on the PATO tracker; see:


Possible solutions

Here is part of a strawman shape hierarchy:

  • shape
    • object shape : inheres in object
      • concave : with open cavities
        • concave lacking holes
          • concave shape with points
            • concave shape with 3 points
              • T-shaped
              • Y-shaped
            • ccv shape with n points
              • star-shaped
        • concave with holes
          • torus
      • convex : lacks open cavities
        • sphericality
    • fiat object part shape : inheres in object part
      • curvature
      • notched

The principle here is to first distinguish between whole shapes and pieces of shapes. For example, "concave" applies to a whole object, but "notched" can apply to a fiat part of that object


This is complementary to whatever other solution is used: PATO terms, especially shape terms, should have an associated image. See

Fallback Solution : WordNet

One solution is to simply give up, abandon the representation of shape in PATO and use wordnet instead. This does not satisfy many of the requirements, but it is not necessarily any worse than the current flat list

Personal tools