RE: Performance issues with OWL Reasoners => subclass vs instance-of

Chris is right, but the IS itself has no view on the matter. it does, 
I believe, play some tricks inside making instances classes to do the 
reasoning. What the user sees are instances. When we use the IS to 
classify proteins, we have a class "p53" and we translate all the 
genes in a genome into  their protein and classify. In this way we 
find instances of the defined proteins classes (in our case 
phosphtases). it is not "real" in that the one  gene  is only 
trnslated into one protein, but it is still an instance, it is simply 
not realistic. We weren't doing syste4m bioogy!

robert.
At 01:12 15/09/2006, Kashyap, Vipul wrote:


> > With InstanceStore, the genes and gene products are treated as owl
> > individuals - belonging to the ABox. However, the ontologically
> > correct representation recognises that p53 is the name of a universal
> > that is instantiated in trillions of cells, and not the name of an
> > individual region of DNA in an individual nucleus, and thus best
> > represented in the TBox. This is how we are thinking of presenting GO
> > anntations in OWL. This is obviously problematic from a practical POV.
> >
> > It seems we need general patterns for transforming certain subsets of
> > TBoxes into ABoxes for the purposes of reasoning. Any thoughts on how
> > this should be done?
> >
> > Chris
>
>[VK] Actually this gets to the heart of one of the design issues identified by
>the BIONT Task Force in designing the Parkinson's Disease Ontology.
>Is a given gene a subclass or an instance-of a general "Gene" class.
>
>http://esw.w3.org/topic/HCLS/OntologyTaskForce/DesignIssues (point number 2)
>
>An initial draft of the ontology is available at:
>
>http://esw.w3.org/topic/HCLS/OntologyTaskForce/SeedOntology
>
>Obviously, if mapping into instances gives better performance for a 
>given set of
>inferences, that might be the basis of choosing the instance-of relationship.
>Towards this end I have the following questions for Phil:
>
>1. What are the set of Abox inferences implemented in the GO example?
>2. What would be the corresponding set of TBox inferences implemented if the
>design choice proposed by Chris was adopted, i.e., p53 is a subclass of Gene
>(assuming a general "Gene" class)
>3. What are the performance and scalability implications of (1) vs (2)
>4. What are the expressiveness implications of (1) vs (2), i.e., can 
>we express
>some statements using subclass-of based modeling which are not possible using
>instance-of modeling; or vice versa....
>
>Look forward to a good use case illustrating the above and discussing its
>possible consequences.
>
>Thanks,
>
>---Vipul

Received on Friday, 15 September 2006 06:35:43 UTC