- From: Ian Horrocks <horrocks@cs.man.ac.uk>
- Date: Tue, 7 Mar 2006 14:40:54 +0000
- To: Jim Hendler <hendler@cs.umd.edu>
- Cc: Ulrike Sattler <Ulrike.Sattler@manchester.ac.uk>, RIF WG <public-rif-wg@w3.org>
On 6 Mar 2006, at 15:12, Jim Hendler wrote: > > At 13:49 +0100 3/6/06, Ulrike Sattler wrote: >> On 3 Mar 2006, at 16:49, Jim Hendler wrote: >> >>> >>> Paul - there's a lot more to this, and some of it is in the >>> vocabulary -- let me start simple - in this case when I say "OWL" I >>> don't mean the theoretical content of some ontology, I mean a >>> document somewhere on the Web which contains RDF/OWL description - >>> for example, if you point your browser at >>> www.mindswap.org/2003/CancerOntology/nciOncology.owl (be careful, it >>> is big) then you would see a particvular version of this ontology >>> (the definitive version lives in NIH space and I don't have the link >>> at the moment, it's considerably bigger than this 25M one). >>> >>> Enter RIF (phase 2) - we want to take the stuff in that NCI >>> ontology, take that document to our cancer research center, and >>> check to see whether the data gathered in our datatbase corresponds >>> to the expectations of the OWL model. So, for example, if we see >>> that an ONCOGENE is defined as any GENE which is ASSOCIATED with a >>> DISEASE that is of type CANCER - we could go to the database and >>> pull out all the things that are ONCOGENEs by this definition, >>> This couldn't be done by OWL itself - it's a variant of the famous >>> "uncle" thing- essentially we need a chaining rule in here (from >>> Gene to disease and disease to cancer) >>> >> >> I am sorry, but I have to disagree: this is easily possible in OWL >> itself: simple retrieve all instances of ONCOGENEs: any DL reasoner >> (eg Pellet) will retrieve exactly all instances of oncogene, whether >> they have been declared as such or whether they are genes that are >> associated with a disease of type cancer.... >> >> Cheers, Uli > > Uli - I'm talking a mix of A-box and T-box with chaining, exactly like > the uncle case, I didn't specify details as I figured it wasn't needed > - but even if I got this case wrong, the point is someone w/a DB may > want to "ground" a set of assertions and then find all the values in > the DB via chaining - so it requires multiple steps of non-functional > predicates - which is what Ian keeps telling us all that OWL cannot do > in its pure form. As Uli says, we can easily answer *extensional queries* that include this kind of property chaining - i.e., we can determine, for any given instance of GENE, if it is a member of ONCOGENE (either as a result of known relationships with instances of DESEASE and CANCER, or because we can infer the existence of such individuals). If we add a "DL Safe" rule of the form: causes(x,z) <- GENE(x) ^ DISEASE(y) ^ associatedWith(x,y) ^ hasType(y,z) ^ Ind(x) ^ Ind(y) ^ Ind(z) where Ind is a predicate that doesn't occur in the ontology and is true for all known individuals, then we can even answer queries that use the causes property (e.g., GENEs that cause CANCER). This does *not*, however, allow us to answer *conceptual queries* using the causes property, i.e., we cannot determine that *any* ONCOGENE (even one we don't know about yet) is necessarily also a GENE which causes CANCER. If we relax the safety condition so that the rule applies to *all* individuals (i.e., even to individuals whose existence is only implicit), then queries would, in general, be undecidable. Ian > -JH > > -- > Professor James Hendler Director > Joint Institute for Knowledge Discovery 301-405-2696 > UMIACS, Univ of Maryland 301-314-9734 (Fax) > College Park, MD 20742 http://www.cs.umd.edu/~hendler > Web Log: http://www.mindswap.org/blog/author/hendler >
Received on Tuesday, 7 March 2006 14:41:04 UTC