Re: [Chebi-ontology] Oxygen-18 to isotope: instance or subclass?

[Alan R. <ala...@gm...>]

I started, last week, to write a detailed response, but have had to
prioritize other work and so hadn't finished. The essential points are:

- The use of instances as classes as suggested by Stefan is not a pattern
we've used, and I'd caution before trying it - it is probably more
mechanism than you need, would require complicating every query involving
types to check both representations, suggests more can be inferred than
actually can be (otherwise the OWL working group would have considered
choosing the representation), and it is likely that your needs, should you
define then (see next point) can be satisfied with a much simpler
representation.

I also want to clarify a couple of things that Stefan said in his note.
Punning is *not* a syntactic feature. Punning doesn't occur in OWL2 using
the RDF Semantics. In that case every class is also an individual. Saying
two individuals are the same makes their class extensions equivalent. Note
that in either semantics, two individuals can be different but have the
same extension. A consequence of this is that in Stefan's proposal, where
you might think saying the types (represented as instances) are different,
this does not make the types disjoint. As an exercise you can try to figure
out how you *would* say that.

Punning in OWL2 using the direct semantics means that the URI can name two
distinct entities that are unrelated (the pun). When using the direct
semantics, if the individuals are made same, the class extensions are not
forced to be the same. In both semantics it can be the case that different
individuals have the same class extensions. The implementation of the
behavior in OWL2 with the direct semantics can be implemented using a
syntactic transformation, which is what Stefan was referring to.

- You need to say what you expect minimal requirements for  query and
inference from a representation with isotopes, since without that we won't
be able to evaluate any representation. My gut is that a simple
representation that uses an annotation property that relates the iris of
the classes would be sufficient under the assumption that you are querying
in SPARQL.   So:

oxygen-16 isIsotopeOf oxygen-17
oxygen-17 isIsotopeOf oxygen-16
oxygen-16 isIsotopeOf oxygen-18
oxygen-18 isIsotopeOf oxygen-16
oxygen-17 isIsotopeOf oxygen-18
oxygen-18 isIsotopeOf oxygen-17

The pairing is because we can't make annotation properties symmetric,
though you could assert the one-way relations and use SPARQL Construct to
add the others.

- You (and perhaps ChEBI) need to review the entire set of representations
related to atoms. To be frank, I'm not sure it is all consistent, in part
because the definitions are not tight enough. But see  oxygen molecular
entity, elemental oxygen, monooxygen, monoatomic oxygen, oxygen atom,
oxide. Then there are all the compounds that have some form of oxygen with
along with other atoms.

I believe the modeling of isotopes needs there to be a representation of
atomic nuclei. All the above classes can then state that they have a part
that is some oxygen nucleus. The isotopes are subclasses. In addition the
annotation relations I suggest would give you a way to relate one isotope
to another. With that representation you would be immune from having to
represent, in each of the listed cases that is disjoint, the same thing
about isotopes. Without it you have to.

Regards,

-Alan



On Mon, Aug 4, 2014 at 8:22 AM, Emw <emw...@gm...> wrote:

> Stefan, Janna,
>
> My original question, i.e. whether the model
>
>
>     oxygen-18
>         *subclass of* oxygen
>         *instance of* isotope
>
>       (where *instance of* is rdf:type and *subclass of* is
> rdfs:subClassOf)
>
> is compatible with ChEBI, seems to depend on whether class-instance
> punning is valid in BFO.  If it is, then this negates the foundational
> definition of "instance" in Smith and Rosse 2004 [7]: "Instances are
> individuals (particulars, tokens) of special sorts... bound to a
> specific... location in space and time."
>
> Is class-instance punning compatible with BFO, and thus ChEBI?
>
> Since this question also concerns BFO, I've copied Alan Ruttenberg on the
> conversation.  The following response explores this further.
>
>
> "Punning" is possible in OWL Full, but it is first of all a syntactical
> feature. Description logics reasoners ignore it, and its semantics is
> undefined.
>
> Punning is possible in OWL 2 DL and supported in description logics
> reasoners like Pellet [1, 2].  The formal semantics of punning are defined
> in Motik 2007 [3], which describes a decidable extension of SROIQ(D) using
> contextual semantics.  Motik 2007 is the basis of the class-instance
> metamodeling supported in OWL 2 DL.
>
> [Metamodeling an entity as a class and an instance, as in the cited
> UniProt example] may be motivated by pragmatic reasons, but is no
> recommendable solution as long as we envisage ontologies as interoperable
> descriptions of segments of the world, and not just as logical models that
> support specific use cases.
>
> Part of me agrees with this.  But how is that recommendation tenable when
> most authoritative explanations of punning in OWL 2 DL use precisely the
> kind metamodeling that UniProt does, i.e. "human *instance of *taxon,
> human* subclass of* primate?  Here is a representative example from
> Horrocks and Patel-Schneider 2010 [4]:
>
> In contrast, OWL 2 allows the same name to be used to refer to different
> types of entity; for example, ex:Eagle might be used to denote both the
> class (a subclass of ex:Bird) and an individual (an instance of ex:Species):
>
>     Class: ex:Eagle
>         SubClassOf: ex:Bird
>     Individual: ex:Eagle
>         Types: ex:Species
>
> (See also the canonical "Harry the Eagle" example used in Motik 2007 [3],
> Grau et al. 2008 [5], Krötzsch 2014 [6], etc.)
>
> This poses a dilemma for those wanting to interoperate in letter and
> spirit with BFO.
>
> The dilemma emerges from the fact that BFO and OWL 2 DL seem to have
> conflicting definitions of "instance".  BFO states that instances are
> particular tokens with a unique location in space and time, while classes
> are universal types [7].  However,  the motivating example for one of OWL 2
> DL's major new features contradicts this: as an instance of "species",
> "Eagle" is *not* a particular thing with a unique location in space and
> time.
>
> Such metaclasses pop up in numerous domains: "human *instance of* taxon,
> human *subclass of* primate"; "Ford Focus *instance of* car model, Ford
> Focus *subclass of* compact car"; "Nimitz-class aircraft carrier *instance
> of* ship class, Nimitz-class aircraft carrier *subclass of* aircraft
> carrier".  These examples are based on classification discussions in
> Wikidata.
>
> I see two options to resolve this dilemma:
>
>    1. Allow class-instance metamodeling in BFO.  This entails broadening
>    the definition of "instance" in BFO to account for classes being
>    instances of metaclasses.  It would clarify that statements like those in
>    the canonical OWL 2 DL examples, UniProt, and the preceding paragraph are
>    all compatible with BFO.  This has a cost in that would, as Stefan
>    puts it, detract from our ability to "envisage ontologies as
>    interoperable descriptions of segments of the world, and not just as
>    logical models that support specific use cases."
>
>    2. Prohibit class-instance metamodeling in BFO.  This would preserve
>    the foundational definition of "instance" in Smith and Rosse 2004 [7]:
>    "Instances are individuals (particulars, tokens) of special sorts... bound
>    to a specific... location in space and time."  However, it would also
>    prevent BFO ontologies from using a major feature in OWL 2 DL.
>
> Stefan notes another possibility:
>
>> If one wants to represent types differently from classes it has been
>> proposed to represent them as OWL individuals.
>>
> However, that would have the same philosophical cost as Option 1 above,
> because -- ontologically, at least in BFO -- classes are types [7].  If we
> use *instance of* to relate two types, then it seems we are treating the
> "instance" and "class" constructs in OWL as just syntactic features with
> no philosophical meaning.
>
> I have searched for literature on how BFO accounts for class-instance
> punning in light of the Smith and Rosse 2004 definition of "instance", but
> have not found anything relevant.  This might be because
> philosophically-oriented BFO literature, e.g. Smith and Ceusters 2010
> [9], tends to predate the publication of OWL 2 in December 2012, which
> brought punning from the controversial OWL Full vocabulary into the
> mainstream OWL 2 DL vocabulary.
>
> Thanks,
> Eric
>
> P.S.: Janna, thanks for correcting my misunderstanding about ChEBI.  I see
> now that chebi.owl [8] defines object properties which have rdfs:domain
> values, but avoids modeling classes as instances by using those properties
> only in OWL restrictions.
> 1.  http://www.w3.org/TR/owl2-new-features/#F12:_Punning
> 2.   "Pellet supports reasoning with the full expressivity of OWL-DL
> (SHOIN(D) in Description Logic jargon) and has been extended to support the
> forthcoming OWL 2 specification (SROIQ(D)), which adds the following
> language constructs: ... vocabulary sharing (punning) between individuals,
> classes, and properties ..."
> http://clarkparsia.com/pellet/features#standard
> 3.  Boris Motik (2007).  *On the Properties of Metamodeling in OWL*.
> https://www.cs.ox.ac.uk/Boris.Motik/pubs/motik07metamodeling-journal.pdf
> 4.  Ian Horrocks and Peter Patel-Schneider (2010). * Knowledge
> Representation and Reasoning on the Semantic Web: OWL*.
> http://www.cs.ox.ac.uk/ian.horrocks/Publications/download/2010/HoPa10a.pdf
> 5.  Bernardo Cuenca Grau et al. (2008). *OWL 2: The Next Step for OWL*.
> http://www.websemanticsjournal.org/index.php/ps/article/viewFile/156/154
> 6.  Markus Krötzsch (2014). *Foundations of Semantic Web Technologies:
> OWL 2 - Syntax and Semantics*.
> http://www.inf.tu-dresden.de/content/institutes/ki/cl/study/summer14/fswt/slides/FSWT2014-L11-OWL2.pdf
>
> 7.  Barry Smith, Cornelius Rosse (2004). *The Role of Foundational
> Relations in the Alignment of Biomedical Ontologies*.
> http://ontology.buffalo.edu/medo/isa.pdf
>  8.  ChEBI OWL export.
> ftp://ftp.ebi.ac.uk/pub/databases/chebi/ontology/chebi.owl
> 9.  Barry Smith, Werner Ceusters (2010). *Ontological Realism as a
> Methodology for Coordinated Evolution of Scientific Ontologies.  *
> http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3104413/
> <http://iospress.metapress.com/content/1551884412214u67/fulltext.pdf>
>
>