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

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

On Tue, Sep 2, 2014 at 11:08 PM, Emw <emw...@gm...> wrote:

> Alan,
>
>
>> I think you first need to have a solid definition of what an isotope is.
>> I offered a start of one from the perspective of BFO in my last email
>> message. I think it would be good for you to offer an alternative. Simply
>> saying the isotope is a metaclass of Oxygen-18 isn't an adequate definition.
>>
>
> You provisionally define isotope as follows:
>
> to be an isotope is something like to be a mutation. To be a mutation is
>> to be slightly different from the canonical, but of equal status except for
>> perhaps the number of class members. To be an instance of an isotope of
>> oxygen (atom) is to be a an atom with a different number of nucleons
>> (protons and neutrons)  than the most abundant form (oxygen-16, with 8
>> protons and 8 neutrons in its nucleus).
>>
>
> I don't think that will work.  Oxygen-16 is an isotope of oxygen and is
> referred to as such throughout the literature.  Any definition of isotope
> must account for that.
>
> A typical textbook defines isotopes as "two atoms of the same element
> having a different number of neutrons" [1].  Wikipedia defines isotope
> similarly [2].  The textbook definition seems workable.
>

This is a relation, not a type. The clue is that it says "two atoms". A
good idea would be to ask what the two atoms are in your statement
"Oxygen-16 is an isotope of oxygen".


> That definition in itself doesn't warrant modeling isotopes as
> metaclasses.
>

Or even classes. What I don't understand is how you move from a binary
relation ("two atoms") to a unary property (implicitly of one atom).


> But criterion D from my 2014-08-27 message [3] arguably could.  Criterion
> D: "Avoid any potential subclasses of the isotopes in (A) and (B) without
> relying on those subclasses being leaves in a subsumption hierarchy."  See
> the criteria atop http://sourceforge.net/p/chebi/mailman/message/32766380/
> for context.
>

Actually I didn't understand that. However I think we should start with the
discordance between considering isotope being a relation vs. isotope being
a class. As a comparison, consider the relation of adjacency. Is there a
class "adjacents"?

Most of what comes below is phrased in terms of the unary interpretation of
isotope. I'll not comment on it further until we sort out this unary versus
binary thing, in the interest of not confusing things.


> Such subclasses of a "particular" isotope are plausible, e.g. carbon-14 in
> fossil fuels, carbon-14 in human bodies.  We could satisfy criterion D by
> using SPARQL queries that do not use inference, e.g.
>
>    - Get *only* classes like oxygen-17, oxygen-18 and oxygen-19:
>
>    SELECT ?subject WHERE { ?subject rdfs:subClassOf oxygen . ?subject
>    rdfs:subClassOf isotope . }
>
>
>    - Get all subclasses of oxygen and isotope:
>
>    SELECT ?subject WHERE { ?subject rdfs:subClassOf* oxygen . ?subject
>    rdfs:subClassOf* isotope . }
>
> (Note the asterisks in the latter.)
>
> That isn't enough, though.  It would still not filter out direct
> subclasses of an element and isotope that are not "particular" isotopes
> like oxygen-18, e.g. "stable isotopes of oxygen", "radioactive isotopes of
> oxygen".  This complication could be addressed by using the statement "*instance
> of* isotope" on "particular" isotopes like oxygen-17, oxygen-18 and
> oxygen-19 but not subclasses of oxygen and isotope like "radioactive
> isotopes of oxygen".  That is the motivation for representing isotope as a
> metaclass.
>
> *is isotope of*, Chris Mungall's proposed BFO-compatible solution, also
> has issues as described in
> http://sourceforge.net/p/chebi/mailman/message/32658963/.  Simply put, it
> is not valid in OWL 2 DL to declare "*is isotope of* rdfs:subPropertyOf
> rdfs:subClassOf",
>
I don't see that as an issue of not being valid OWL 2 DL. To me it is an
issue of whether it makes sense. To me it doesn't.

Here is another question. How would you distinguish the class isotope from
the class atom?

> and the statement we'd thus be forced to make -- "oxygen-18 *subclass of*
> oxygen and oxygen-18 *is isotope of* oxygen" -- is awkwardly redundant.
>
Here again, even in the relational expression, the textbook definition
can't be understood unless you say what the members of the class oxygen
are. If the members of the class are all atoms with 8 protons in the
nucleus, then in the above you would have instances of oxygen-18 being
isotopes of oxygen-18.

Aside from that, it isn't redundant - one of the axioms describes class
membership and the other expresses a relationship between members of the
class. The criteria of being awkward or not doesn't, to me, seem to be a
relevant criteria at this point. First you have to make clear what you
mean. Then we can discuss alternatives to representing that.

> So none of the options here are good.  Option A -- "oxygen-18 *subclass
> of* oxygen and *subclass of* isotope" -- is compatible with BFO and ChEBI
> but requires brittle conventions to meet the querying and inference
> requirements in [3]. Option B -- "oxygen-18 *subclass of *oxygen and
> oxygen-18 *instance of* oxygen" is incompatible with BFO and thus ChEBI.
> Option C -- "oxygen-18 *isotope of* oxygen and *isotope of*
> rdfs:subPropertyOf *subclass of*" or "oxygen-18 *subclass of* oxygen and *isotope
> of* oxygen" is either incompatible with OWL or awkwardly redundant.
>
> I see no way forward.  How can oxygen-18 be modeled as an isotope in a way
> that is compatible with BFO and OWL 2 DL, and also meets the criteria in
> [3]?
>

The way forward is to first get clear on what the ontology of isotope is,
independent of OWL. Then we can work on OWL.


>
> You are free to use BFO or not.
>>
>
> Wikidata is certainly free to use whatever upper ontology it desires.  But
> I think compatibility among Wikipedia's ontology and BFO, ChEBI and OBO
> would be a major boon to the Semantic Web.
>

I would tend to agree. One of the benefits of using BFO, in my view, is
that it tries to force the user to be clear about what they are talking
about. So let's work on that goal.


>
> Currently, Wikidata is not compatible with BFO or ChEBI.  There is a
> tendency in the Wikidata community to use *instance of* in subjects that
> are clearly classes, i.e. universals -- e.g. carbon *instance of*
> chemical element, oxygen-18 *instance of* isotope, heart attack *instance
> of* cardiovascular disease [4-6].   As we have established here, such
> statements are not compatible or consistent with BFO, ChEBI and other OBO
> ontologies.
>

More importantly, it isn't clear what they mean.


>  But they do offer an easy way to solve important querying tasks -- like
> "how do I get a list of (elements | isotopes | diseases)?" -- even though
> they contradict BFO's definition of instance as something with a particular
> location in space and time.
>

My experience has been that by focusing first on ontology, afterwards the
query problem tends to look much easier.

>
> I look forward to any more proposals on how to move through this impasse!
>

Start with the two problems I have identified

1) Discordance between isotope as relation versus class. Use
adjacent-to/adjancets as an analogy to help guide your thinking.

2) Even in the (textbook) relational view, think about whether oxygen-18 *is
isotope of* oxygen is a sensible statement, or whether only statements of
the sort oxygen-16 is isotope of oxygen-17 are the right sort of
statements. I think you might gain some clarity by putting a critical eye
on some of the problems with the definitions one might identify as an
ontologist. Here I discuss the textbook definition, but the same problems
are present in the wikipedia definition.

original "two atoms of the same element having a different number of
neutrons"

This sounds like an isotope is an aggregate of two atoms, which of course
is not what is meant, despite that being the most direct interpretation of
the english.

Some alternatives:

"the relation between two atoms that are instances of the same type of
element but which have a different number of neutrons"

With the guideline that types of elements are classes all of whose members
are atoms with the same number of protons in their nucleus.

or

"the relation between two types of atoms, with instances of both having the
same number of protons in their nuclei and the differentia of the types
being that respective instances differ in the numbers of neutrons"

While these definition are perhaps not in the vernacular of the chemist,
the explicit use of type/class/universal/relation in the definitions
removes ambiguities (and therefore confusion) present in the natural
english definitions.


-Alan


> Best,
> Eric
>
> 1.  Janice G. Smith (2006).  Organic Chemistry, 1st edition.  Page 7.
> 2.  Isotope on Wikipedia.
> https://en.wikipedia.org/w/index.php?title=Isotope&oldid=623706639
> 3.  http://sourceforge.net/p/chebi/mailman/message/32766380/
> 4. Carbon on Wikidata.
> https://www.wikidata.org/w/index.php?title=Q623&oldid=151858101
> 5. Oxygen-18 on Wikidata.
> https://www.wikidata.org/w/index.php?title=Q662269&oldid=84117617
> 6. Heart attack on Wikidata.
> https://www.wikidata.org/w/index.php?title=Q12152&oldid=155099715
>