Re: OWL vs RDF

Hi All,

At this week's BIRN All-Hands Meeting, the BIRN Ontology Task Force  
has been working on ontological representation of neuroanatomy  
(decomposing the differing senses often wrapped together in a  
standard neuroanatomical terminological resources), so that in the  
end we'll be able to recompose an OWL repsentation of the  
traditionally convolved dualistic structure-function description of  
neuroanatomy.  If this process proves as promising as it appears to  
be so far, we'll hopefully be able to include the additional contexts  
we know we also need to take into account to represent those entities  
and properties that are also ubiquitous in traditional  
neuroanatomical studies, such as cytoarchitectural detail,  
neurochemistry, development, cross-species comparative anatomy, gene  
& protein expression, etc.

I would add the work we are doing in neuroanatomy relies heavily on  
existing knowledge resources in this domain which we've found we  
cannot use in their current form.  As we've proceeded with the work  
above over the last year, we've worked hard to stay aligned with  
these resources as much as is practical.

We've also been working on the standard cognitive and behavioral  
paradigms used now - in the context of the BIRN and many other  
projects - as an experimental context for functional brain imaging  
experiments.  This also requires careful decomposition and  
examination of how to properly create object properties, so that many  
of the more complex aspects of the ontological graph can be  
assembled, extended, and maintained via reliance on the DIG reasoners  
that can be used in ProtegeOWL, for instance.

It's pretty clear if we don't proceed by careful consideration of how  
and when to lean on the reasoners, the issues Robert and Phil discuss  
below - and others have been discussing in the context of the NCBO/ 
OBO Foundry-associated community biomedical ontology development  
projects that are on going - we will very quickly overwhelm the  
limited human resources we have in place to both build, extend, and  
maintain these ontologies.

In both contexts cited above - neuroanatomy and cognitive &  
behavioral experimental psych. paradigms, we are already working on  
how to build from the tools being assembled by the community - BFO,  
CARO, Relations Ontology, etc. - and others of our own creation - the  
Object Properties we'll need to re-compose a complete ontological  
representation of these complex domains.

This is not to say this work cannot be done using RDF alone, but I  
don't see how we'll be able to get away without relying on formal  
rules for constructing constraints and complex, qualified ontological  
relations.

In the end, I suppose the question really will come down to whether  
the limits, entailments, and computational constraints one can  
confront when using OWL (specifically OWL-DL) will prove to be more  
of a hindrance than the additional work required to construct such  
ontological frameworks in RDF alone.

As Phil implies below, if you really just need a formalism for nodes  
and edges, you don't NEED OWL - in fact, for many types of graphs,  
you don't need RDF.  On the other hand, if you are clearly going to  
require a formalism with considerable ontological expressivity, you  
probably want to give the OWL dialects (and their underlying DLs and  
the toolsets such as ProtegeOWL  and the Pellet, FACT++, and other  
reasoners, etc.) serious and in depth consideration.

Just to be clear, I'm talking here about ontology construction which  
I consider just a portion of the required task of semantically formal  
data representation.  For much of the semantic representation we need  
to do in large scale biological data repositories, RDF alone will  
clearly be a sufficient first step, so long as we continue to develop  
effective means of expressing the triplets in the context of the  
ontologies and extending the ontologies via analysis (as automatic as  
is feasible) of the triplet repositories.

Cheers,
Bill

On Oct 26, 2006, at 1:45 PM, Robert Stevens wrote:

>
> There's another answer of using the reasoner by building your  
> ontology to take advantage of its capabilities. the conceptual lego  
> approach relies on the reasoner.
>
> Anecdotally, I've managed to miss out subsumption relationships by  
> hand in ontologies as small as a dozen classes. Also, when building  
> purely by hand, I've seen people introduce inconsistencies by  
> asserting multiple inheritance across disjoint classes.
>
> I would feel nervous at saying "there is a point when you need a  
> reasoner", but I think building to take advantage of it is as close  
> as I can get.
>
> robert.
> At 11:46 26/10/2006, Phillip Lord wrote:
>
>> >>>>> "Alan" == Alan Ruttenberg <alanruttenberg@gmail.com> writes:
>>
>>   Alan> Well it would be educational to get your view on what you can
>>   Alan> you do with owl without a reasoner that's not easier to do
>>   Alan> without owl?
>>
>>
>> You can do lots of things with OWL without a reasoner. The Gene
>> Ontology is representable in OWL, for example, and uses a simple
>> enough expressivity that you could do without a reasoner easily
>> enough. Of course, you need to use some kind of "reasoning" engine,
>> but something which understands transitive closure is enough.
>>
>> Whether it's "easier" to do without OWL depends on what the
>> alternatives are. You could also represent GO style semantics in RDF
>> (although, I think, the existential nature of part_of would not be
>> explicit), or indeed anything else capable of representing a
>> graph.
>>
>>   Alan> And how are you to know when you do need the reasoner and  
>> when
>>   Alan> you don't?
>>
>> When you use enough of the expressivity of OWL, where "enough" is
>> relatively undefined.
>>
>> Phil
>
>
>

Bill Bug
Senior Research Analyst/Ontological Engineer

Laboratory for Bioimaging  & Anatomical Informatics
www.neuroterrain.org
Department of Neurobiology & Anatomy
Drexel University College of Medicine
2900 Queen Lane
Philadelphia, PA    19129
215 991 8430 (ph)
610 457 0443 (mobile)
215 843 9367 (fax)


Please Note: I now have a new email - William.Bug@DrexelMed.edu