[MOBY-l] DAML/OIl editor for viewing MyGrid ontology

[Phillip Lord]

>>>>> "Simon" == Simon Twigger <simont at mcw.edu> writes:

  Simon> Hi Phil,

  Simon> On Apr 5, 2004, at 12:22 PM, Phillip Lord wrote:
  >>
  >>
  >> Something that I would be keen to do is to come up with an OBO
  >> style ontology from the original mygrid one. Mark and I have
  >> discussed this before, but we've never got around to it.

  Simon> By OBO style do you mean convert the existing DAML version
  Simon> into a format accepted by OBO, presumably OWL?

Yes and no. Translating to OWL by itself is (nearly) trivial and
something that we can just do. 

The deeper issue is considering the semantics of OWL and the less
defined semantics of OBO. We could just take the ontology as stands
and turn it into a DAG. But if we just take the class hierarchy we
will loose much of the information. 

  Simon> The Reasoned version of the DAML makes a huge difference -
  Simon> thanks.

  Simon> Just to check my understanding of what's going on here -
  Simon> reasoning the ontology has pulled together all possible
  Simon> hierarchies that were defined in the model but previously
  Simon> 'hidden' in the unreasoned version by virtue of it being a
  Simon> non-redundant listing of terms? Hence the reasoned version
  Simon> now shows every allowable path to a class and so you get a
  Simon> better perspective of what lives where?


Not really no. The hierarchy browser in Oiled shows the information
that it has. That terms occur redundantly is just a way of displaying
a multiple inheritance hierarchy. 

The reasoner does something more complex. In a description logic based
ontology you tend to model the classes by describing their
properties and then let the reasoner work out what subclass and super
class relationships. 

Take the following example. This is using the (poorly named) OWL
concrete abstract syntax. 
 

  Class(SegregatingUnit  complete
       restriction(involvedIn  someValuesFrom (Segregation )))

segregating units are involved in segregation....

  Class(Chromosome  complete
       intersectionOf(SegregatingUnit
            restriction(hasPart  someValuesFrom (Telomere ))
            restriction(hasPart  someValuesFrom (Centromere ))))

chromosomes are things which are segregating units, and have a part
which is a telomere and a part which is a centromere. This is
definitional. 


  Class(CloningVector  partial
       SegregatingUnit )

cloning vectors are segregating units also...

  Class(YAC  partial CloningVector
      restriction(hasPart  someValuesFrom (Telomere ))
      restriction(hasPart  someValuesFrom (Centromere )))

A YAC (yeast artificial chromosome...all the rage when I still worked
in the lab!) is a cloning vector with a telomere and centromere. 

>From which we can conclude that a YAC is a chromosome as it's a
segregating unit with a telomere and a centromere. 

We can also conclude that...

  Class(AcentricChromosome  partial Chromosome )
  DisjointClasses(restriction(hasPart  someValuesFrom (Centromere ))
      AcentricChromosome )


a chromosome with out a centromere is a contradiction in terms....


  DisjointClasses(AcentricChromosome
             SegregatingUnit)   


as is a chromosome which is not a segregating unit. 

In the mygrid ontology a lot of the super class/sub class links are
not stated explicitly but only generated by the reasoner. This style
of modelling is sort of like Object orientation...but upside
down. Properties define super classes, rather than sub classes
inheriting properties. 

This is one of the big selling points of OWL (or at least OWL-DL). Its
highly expressive but still computationally tractable. We can
mathematically guarantee that any statement you can make in OWL-DL can
be reasoner over. This is not true of OWL-Full or RDF; there are
questions which can not be answered: that is are "undecidable". If
this sounds counter intuitive then it can be considered as a variation
on the theme of Turing's halting problem. There are somethings you
just can't work out. 

Anyway from a practical point of view the reasoned ontology has all of
the super/sub class relationships explicitly stated. It's the
"compiled" form if you like. 

Cheers

Phil