Search Mailing List Archives


Limit search to: Subject & Body Subject Author
Sort by: Reverse Sort
Limit to: All This Week Last Week This Month Last Month
Select Date Range     through    

cell component ontology

Peter Karp pkarp at ai.sri.com
Wed Nov 2 05:49:04 PST 2005


Dear Wacek:

I have to disagree with several points in your message.

Waclaw Kusnierczyk wrote:
> Dear Peifen (and Others),
> 
> Actually, Anand's complaint does not seem to be ungrounded at all, and I 
> feel his point has to be supported.
> 
> If your criterion to distinguish between a class and an instance is that 
> you simply do not have a further specialization for (what you call) an 
> instance, this is a misconception. An instance is (the representation 
> of, the pointer to) a particular object in a domain; 

The critical point here is what is meant by "a particular object in a
domain"?  This issue comes up in virtually all semantic modeling
exercises, in which there are frequently a large range of choices as to
what is meant by a "particular object", depending on how fine are
the distinctions we choose to make in our model.  For example, the
"particular object" meant by "nucleus" could be interpreted in many
possible ways.  It could refer to a nucleus of a particular cell at a
particular moment in time, the nucleus of that same particular cell
but at all points in time, the nuclei of all members of a population
of identical cells (should we choose to treat them as a single entity
for the purposes of our model), or the nuclei of all members of
a population of non-identical cells (should we choose to treat that
heterogeneous population as a single entity for the purposes of our
model).  Our instance "nucleus" refers to whatever cell or population
of cells the user of our ontology is choosing to treat as a single
entity (an instance) in their model.

Yes, of course we could have chosen to make "nucleus" be a class
instead of an instance, however, for the purposes for which this
ontology was designed, we wished to emphasize in our model the
notion of cellular structures as the base-level objects of discourse.
You will note that our ontology is consistent in this sense.

My point, again, is that there is rarely if ever an objective
definition of what is "a particular object in a domain" --
it is virtually always a function of the modeler's perspective,
and the uses for which the ontology is designed.

 > a class is an
> abstract description of a group (possibly empty) of objects. That is, an 
> instance really refers to one and only one object, while a class refers 
> to an arbitrary number of objects that share some properties (which the 
> class describes, or, if you prefer, which form the description of the 
> class).
> So nucleus and organelles in your 'ontology' (to be discussed later) are 
> in fact classes, since they do not correspond to 'that nucleus in that 
> single bacterium', as Anand points out, but rather refer to nuclei etc. 
> in general.

Hopefully you now understand why this point is incorrect.  It is the
modeler who chooses what they consider to be an instance in their
ontology.
> 
> Another point is that a class is an abstraction of objects in the 
> domain, and a particular class may be a generalization of another class. 
> Thus, ER is an abstraction of all objects that you might classify as ER, 
> and a generalization (here you are correct) of two subtypes of ER which 
> are (here you are wrong) also classes, that are abstractions of objects 
> that you might classifiy as either rough ER or smooth ER.

I don't see a separate point here; the same class/instance distinctions
apply.
> 
> Another issue is that an ontology is a conceptual schema about a domain, 
> a specification of what the fundamental categories (and the relations 
> between these) of objects there are in a domain. An ontology does not 
> contain instances. An ontology describes instances via their 
> abstractions (that is classes, concepts), but does not name (point to) 
> instances.
> Here, again, you introduce a conceptual misconception.

Again you are incorrect.  Indeed, ontologies are conceptual schema about
a domain, but often in the course of defining such conceptual schemata
it is critical to include definitions of instances as well as classes.
Particularly in their roles as communicating shared vocabularies, it is
often necessary for them to establish vocabulary terms that refer to
well-known instances in the domain.
> 
> Gene Ontology, since it was introduced a few years ago, has been 
> extensively used (though it is arguable whether it has really been used, 
> or it is just trendy to refer to), yet it has been shown to have major 
> misconceptions, which may cause trouble when computer scientists attempt 
> to use it in an advanced, automated, knowledge-representation-and-
> reasoning based way. It is agreed that GO needs conceptual corrections 
> to become a real ontology. The point is that any new 'ontology' that is 
> being introduced should be carefully crafted, mindfully thought of 
> before being published. It may seem just a scholar discussion of what is 
> a concept and what is an object, but at the implementation level of a 
> reasoner you really need to know.
> 
> A suggestion is to have a look at some reference on the subject, 
> starting from the lightweight introduction in wikipedia or the like, 
> finishing on texts in philosophy and metaphysics.
> 
> Regards,
> Wacek
> 
> 
> 
> 
> 
> On Tue, 1 Nov 2005, Peifen Zhang wrote:
> 
>> Dear Anand,
>>
>> In CCO, we treat ER as a class since it refers to the generalization
>> of the two subtypes of ER, smooth and rough. Nucleus and similarly 
>> other organelles are consistently described as
>> instances throughout our ontology, because they refer to the actual
>> organelles of the organism being modeled in a given database.
>>
>>
>> Best Regards,
>>
>> Peifen
>>
>>
>> On Mon, 31 Oct 2005, Anand Kumar wrote:
>>
>>> Dear Peifen,
>>>
>>> I am not sure which mailing list this should go to. However, you have 
>>> the
>>> following relation in the new ontology.
>>>
>>> Superclasses: organelle
>>> Subclasses: vacuole, microbody, endoplasmic reticulum, endosome, 
>>> vesicle,
>>> plastid
>>>
>>> Instances: thylakoid (sensu Cyanobacteria), nucleus, mitochondrion, 
>>> lysosome,
>>> Golgi apparatus
>>>
>>> GO does not have instances but atleast it does not do the mistake of 
>>> putting
>>> together "nucleus", which is a class, as an instance, which is "this 
>>> nucleus",
>>> "nuclues of that bacterium in that petri dish".
>>>
>>> Why would you treat nucleus as instance and endoplasmic reticulum as 
>>> a class? GO
>>> has problems but atleast not of those sorts.
>>>
>>> Kind Regards,
>>> Anand.
>>>
>>>
>>
>> -- 
>> This message is from the GOFriends moderated mailing list.  A list of 
>> public
>> announcements and discussion of the Gene Ontology (GO) project.
>> Problems with the list?           E-mail: 
>> owner-gofriends at geneontology.org
>> Subscribing   send   "subscribe"   to   
>> gofriends-request at geneontology.org
>> Unsubscribing send   "unsubscribe"  to  
>> gofriends-request at geneontology.org
>> Web:          http://www.geneontology.org/
>>
> 

--
This message is from the GOFriends moderated mailing list.  A list of public
announcements and discussion of the Gene Ontology (GO) project.
Problems with the list?           E-mail: owner-gofriends at geneontology.org
Subscribing   send   "subscribe"   to   gofriends-request at geneontology.org
Unsubscribing send   "unsubscribe"  to  gofriends-request at geneontology.org
Web:          http://www.geneontology.org/



More information about the go-friends mailing list