Difference between revisions of "Hypodigm"

From phenoscape
 
(Problem)
Line 1: Line 1:
 
==Problem==
 
==Problem==
How to handle the uncertainty of tying phenotype annotations to specimens within publications has been at issue.  Our existing data model held all phenotype annotations at the level of particular specimens.  Handling ambiguity at this level was causing difficulty in determining the proper OBO semantics, since authors often do not explicitly reference specimens, but instead discuss taxonomic groups.  This level of detail raises questions of what our goals should be: trying to model the logic within the published descriptions exactly, or rather creating a knowledge base of phenotypic diversity which uses the publications for support.
+
How to handle the uncertainty of tying phenotype annotations to specimens within publications has been at issue.  Our existing data model held all phenotype annotations at the level of particular specimens.  Handling ambiguity at this level was causing difficulty in determining the proper OBO semantics, since authors often do not explicitly reference specimens, but instead discuss taxonomic groups.  Since a list of specimens is included in the paper, we know that at least one of them, and maybe more, were examined for the described character.  This "someOf" semantics is not currently available within OBO and may not be logically desirable.
 +
 
 +
This level of detail raises questions of what our goals should be: trying to model the logic within the published descriptions exactly, or rather creating a knowledge base of phenotypic diversity which uses the publications for support.
  
 
==Resolution==
 
==Resolution==

Revision as of 21:35, 15 November 2007

Problem

How to handle the uncertainty of tying phenotype annotations to specimens within publications has been at issue. Our existing data model held all phenotype annotations at the level of particular specimens. Handling ambiguity at this level was causing difficulty in determining the proper OBO semantics, since authors often do not explicitly reference specimens, but instead discuss taxonomic groups. Since a list of specimens is included in the paper, we know that at least one of them, and maybe more, were examined for the described character. This "someOf" semantics is not currently available within OBO and may not be logically desirable.

This level of detail raises questions of what our goals should be: trying to model the logic within the published descriptions exactly, or rather creating a knowledge base of phenotypic diversity which uses the publications for support.

Resolution

Our discussion at the Biocuration meeting led to a new data model in which all phenotype annotations are expressed as properties of species. Any annotation is assigned an evidence code such as those used by the GO project. An phenotypic description explicitly tied to a specimen will be given a "strong" evidence code, such as "direct observation by author", requiring a supplemental museum catalog number or numbers. A phenotypic description in which the author just says, "species X has phenotype Y", without pointing to a specimen, will be curated as well, but the evidence code will be something like "traceable author statement", and no catalog numbers attached (even though some for that species may be listed in the materials and methods).

The same methodology can be extended to statements about higher taxa. In this case, individual species-level phenotype annotations should be generated for every species of that taxon listed in the materials list. These will generally all get the weaker evidence code. If the author happened to point out one or more particular specimens illustrating the phenotype, these particular species can be given the stronger evidence code with catalog numbers attached.

We need to standardize the list of evidence codes and their meaning for our data. Here is a GO page about them, and here is an evidence_code ontology, with many more codes: