[MOBY-l] ontology

[Catherine Letondal]

Hi,

a.garcia at imb.uq.edu.au wrote:
>    About Mark's email regarding an ontology for analytical 
> methods in bioinformatics. I am here presenting some 
> comments, hope they are useful.
> 
> An ontology for analytical methods in bioinformatics 
>     should ideally allow the discovery of web services from 
>     the conceptual domain and not from the operational domain; 
>     the same being true for those cases where users wish to 
>     define workflows. Ideally the users should be able to 
>     define an analysis, independently of whether this is a 
>     single step or a workflow; a simple example would be 
>     clustal/puzzle clustal/ednadist/ Eneigbor// Econsense 
>     clustal/< > .These are three different cases of 
>     estimating and realizing phylogenies by executing the 
>     three main phylogenetic methods: distance, parsimony and 
>     maximum likelihood. They may be semantically  equivalent, 

They are not semantically equivalent *at all*, but they can be classified together.

>     but are syntactically different. Some of the reasons for 
>     having an ontology are explained at 
>     http://protege.stanford.edu/publications/ontology_development/ontology101.pdf
>     
>     The role of the ontology should be to effectively  identify 
>     these kinds of scenarios. Bioinformatic analytical methods 
>     should be classified according to their conceptual value. 
>     Classes (concepts) are a part of the ontology, properties 
>     for each concept describing various features and 
>     attributes of the concepts (slots, sometimes called roles 
>     or properties). A good example is given in the Protege 
>     manual. In my view concepts are terms such as alignment, 
>     from the properties (slots) the ontology should enable an 
>     identification or at least provide guidance regarding 
>     which alignment method to use depending on the case 
>     (input). 
>     
>     Perhaps I misunderstand the structure presented in the 
>     email, but following that structure may lead to an 
>     ambiguous final result. Is it necessary to include 
>     concepts such as: protein database (is-a)      + 
> databases connecting other databases (is-a)? This ontology 
> should be about analytical methods so is it necessary to 
> include databases? Perhaps those are two different 
> domains. Analytical methods are one thing and databases 
> are another different thing, aren't they? This structure 
> slightly resembles the taxonomy of analytical methods over 
> which EMBOSS relies. But a taxonomy is not an ontology. 
>     
>     Analytical methods are always implementations aren't they? 
>     I mean if they are not implemented then they are not 
>     available as web services. Perhaps the case of multiple 
>     implementations like BLAST (described by the author of the 
>     email) may be resolved by using the concept of the 
>     implementation as the class, for those cases where the 
>     same implementation has several concepts slots should  be 
>     able to resolve the ambiguity. 
>     
>     I am/was working on an ontology for analytical methods, 
>     I am working with Protegue, and basically I am using 
>     bio-catalog as a starting point. The key issue here is 
>     that biomoby requires an ontology for web services,  which 
>     is different from an ontology for analytical methods  (I 
>     guess). 

What is the difference?


--
Catherine Letondal -- Pasteur Institute Computing Center