Difference between revisions of "Modeling Absence of Phenotypes"

From phenoscape
Jump to: navigation, search
(No difference)

Revision as of 05:13, 27 November 2009

Media:PhenotypesInPhenoscape.ppt

Hilmar Lapp

Cartik, don't you think that there is a problem with the lacks_part("process of basihyal bone") class? What is the genus here?

When using inheres_in for the phenotype composition, such as inheres_in("flat", "basihyal bone"), there is a genus ("flat"), a relationship (inheres_in), and a differentia ("basihyal bone") term.

So if you used, for example, lacks_part("body", "process of basihyal bone"), then by inference:

lacks_part("body", "process of basihyal bone") is_a "body" lacks_part("body", "process of basihyal bone") lacks_part "process of basihyal bone" lacks_part("body", "process of basihyal bone") lacks_part_of "basihyal bone"

I don't follow how you infer from lacks_part("process of basihyal bone") that

lacks_part("process of basihyal bone") is_a "absent"

Even if I add a genus term such as in

lacks_part("body", "process of basihyal bone") is_a "absent"

I'm not following how you arrive at that. But do we even need to infer any of that?

Chris Mungall

Cartik's analysis is largely correct and is why we recommend use of relational qualities to indicate an increase, decrease of number of parts (including the limit case, decrease to zero).

http://www.bioontology.org/wiki/index.php/PATO:Revised_2008#Absence_and_counting

Additional comments below

On Aug 6, 2009, at 6:41 PM, Hilmar Lapp wrote:

> > Cartik, don't you think that there is a problem with the > > lacks_part("process of basihyal bone") class? What is the genus here?

I think we need to be more consistent with notation

R(X,Y) typically indicates a single proposition (aka statement, sentence) involving a binary relation

In Cartik's slides he uses: inheres_in(Q,E)

not to denote a proposition, but to denote a class expression: [Q and inheres_in E]

lacks_part is a (binary) relation in ro_proposed. It also exists in PATO as a relational quality or reified relation such that we can talk directly of the lack.

I will choose to interpret lacks_part(X) as the class expression:

PATO:lacks_all_parts_of_type and towards X and inheres_in organism

Or equivalently

E=organism Q=lacks_all_parts_of_type E2=X

> > When using inheres_in for the phenotype composition, such as > > inheres_in("flat", "basihyal bone"), there is a genus ("flat"), a > > relationship (inheres_in), and a differentia ("basihyal bone") term.

Yes

(minor point: the differentia is the relation-relatum pair <inheres_in basihyal_bone>)

> > So if you used, for example, lacks_part("body", "process of basihyal > > bone"),

OK, here is where the notation becomes confusing. This is a proposition stating 'all bodies lack a process of basihyal bone', which is untrue

I will treat this as

PATO:lacks_all_parts_of_type and towards process-of-basihyal-bone and inheres_in body

> > then by inference: > > > > lacks_part("body", "process of basihyal bone") is_a "body"

inheres_in body

> > lacks_part("body", "process of basihyal bone") lacks_part "process of > > basihyal bone"

towards basihyal-bone

> > lacks_part("body", "process of basihyal bone") lacks_part_of "basihyal > > bone"

towards (anatomical_entity and part_of basihyal_bone)

> > I don't follow how you infer from lacks_part("process of basihyal > > bone") that > > > > lacks_part("process of basihyal bone") is_a "absent" > > > > Even if I add a genus term such as in > > > > lacks_part("body", "process of basihyal bone") is_a "absent" > > > > I'm not following how you arrive at that. But do we even need to infer > > any of that?

My reading of the situation is as follows

From the proposition

[1] L-cornutus exhibits absence of (all) proc-of-basihyal-bone

we would like to infer

[2] L-cornutus exhibits absence of SOME process [3] L-cornutus exhibits absence of SOME part of the basihyal-bone [4] L-cornutus exhibits decreased number of processes [5] L-cornutus exhibits decreased number of parts of the basihyal bone

but not

[2x] L-cornutus exhibits absence of (all) processes [3x] L-cornutus exhibits absence of (all) parts of the basihyal-bone [4x] L-cornutus exhibits absence of (all) basihyal-bone [5x] L-cornutus exhibits decreased number of basihyal-bone

2-5 are all correct, 2x-5x are all wrong.

I use 'absence of' in the above propositions for readability, but these would be translated to relational quality form.

Logically, absence is a non-quality cannot inhere in anything, and if you attempt to treat like a quality that stands in an all-some inheres_in relationship you will end up with false inferences such as [2x].

To get the correct inferences, it's necessary to - convert <absent inheres_in X> to the unambiguous relational qualities (preferably as far upstream as possible); you can use the reciprocal_of relation in PATO to help - not treat 'towards' as an all-some relation - add extra horn rules as sql views for these

For the phenotype analysis of OMIM we took a simplified approach and just treated towards as an all-some relation, which made the propositions weaker which we deemed to be ok for statistical matching