[Obo-taxonomy] Questions about your 2008 taxonomy paper

[Peter Midford]

Stefan,

             I've recently re-read your 2008 paper ("The Ontology of Biological Taxa") and it was suggested to me that I ask you directly about a couple of difficulties I encountered.  The first question relates to sections 3.1, and 3.2.  In particular these sections don't seem to argue against what I would consider a fairly standard treatment of ranks as metaclasses:

instance_of(Elephas maximus,Species)

Is_a(Elephas maximus, Elephas)

instance_of("Clyde",Elephas maximus).

I understand that there are some issues with doing this in OWL-DL (or maybe just the 1.0 version), but otherwise it wasn't clear why this approach wasn't explicitly discussed, since it seems to come up in almost every discussion of representing taxonomic ranks I've encountered.

The second question relates to your extension of taxon qualities to organism parts, a laudable goal, but it seems your axiomatization does more than it should.  In some cases, for example:

derives_from(x,y) ^ inheres_in(z,y) -> inheres_in(z,x)

no harm seems to be done if quality z is strictly a taxon quality (e.g., you don't change your taxon quality as you develop from an embryo, though lots of other qualities change).  But the corresponding axiomatization for combining part_of with inheres_in leads to problems even with z restricted to taxon qualities, for example it doesn't seem reasonable to assign Homo sapiens quality to the quarks in a carbon atom in one of my heart cells.  

Your implementation section (4) adds an axiom that ties TaxonQualities to things that contain NucleicAcids.  Although this solves the quark problem, BioTop seems to lack a mereology of NucleicAcids or an axiom indicating what the smallest portion of a NucleicAcid is that is still considered a NucleicAcid - a triplet, a single base?  Limiting this to full chromosomes would deny any taxonomic quality to a species characteristic expressed coding sequence (e.g., a sequence recognizable as human hemoglobin-A), but there seems to be nothing in BioTop that addresses this question.  Also, as Chris Mungall pointed out to me, the taxon quality is not specifically related to a nucleic acid, it inheres_in anything that contains a NucleicAcid.  I had interpreted this differently, specifically that a NucleicAcid bore a TaxonQuality by virtue of, for example, including a taxon specific sequence.  An approach that tied taxon qualities to sequences is easier to interpret, as it corresponds to the approach taken by the DNA barcodes people, which regardless of any empirical problems, is conceptually straightforward.  I'm less certain how to interpret what you did present. 

The containing NucleicAcid requirement also causes another problem for any taxonomy that includes fossils (or potentially non-cellular structures such as hair).  The composition of derives_from and inheres_in you gave in section 3 would reasonably allow assignment of taxonomic qualities to fossils.  However the additional axiom requiring an entity to include NucleicAcid as a part would prevent assigning most fossils to a particular taxon, or any taxon at all. 

I would be very interested in your responses to these difficulties.

Thanks,

Peter

Peter E.  Midford

Phenoscape Ontology Curator

Peter....@gmail.com

[Stefan Schulz]

Dear Peter,

thanks for your interest!

2010/3/14 Peter Midford <peter....@gmail.com>:

> Stefan,
>              I've recently re-read your 2008 paper ("The Ontology of
> Biological Taxa") and it was suggested to me that I ask you directly about a
> couple of difficulties I encountered.  The first question relates to
> sections 3.1, and 3.2.  In particular these sections don't seem to argue
> against what I would consider a fairly standard treatment of ranks as
> metaclasses:
> instance_of(Elephas maximus,Species)
> Is_a(Elephas maximus, Elephas)
> instance_of("Clyde",Elephas maximus).
> I understand that there are some issues with doing this in OWL-DL (or maybe
> just the 1.0 version), but otherwise it wasn't clear why this approach
> wasn't explicitly discussed, since it seems to come up in almost every
> discussion of representing taxonomic ranks I've encountered.

I have several objections against metaclasses:

First, a pragmatic one: if you allow metaclasses, then you get even
more arbitrariness in the ontology engineering process, because each
ontology engineer will have to decide where to put the class /
metaclass boundary, and I do not expect much agreement here. In a
current working group on "owl-izing" the Foundational Model of Anatomy
I am part of, there was a general agreement to get rid of metaclasses
and not to use "punning" (although the group is quite heterogeneous).

Second, a technical one: DL classifiers cannot deal with metaclasses,
and it seems that nobody wants to open Pandora's box and allow "sets
of sets" (Russel's paradox)

Third, a philosophical one (but here you may disagree): I could put it
in a way, that I do not see metaclasses in scientific reality (but I
am no philosopher and I am not dogmatic ;-) ) .
What looks like metaclasses on a first sight could also be interpreted
as information artifacts.
"Elefant is a species" may be semantically interpreted as follows: the
taxonomic trees are information artifacts, consisting of links and
nodes. For the node "Elephant_node" we can say that it denotes only
elephants.
(There is a similar approach to the representation of chemical
entities and chemical graphs: Hastings & Bachelor & Schulz: What are
chemical structures and their relations? Forthcoming in FOIS 2010.

Good point. My favorite example is insulin: Does human insulin bear
the taxon quality human, and does pig insulin bear the taxon quality
pig?
But I think, that the whole discussion boils down to the notion of
derivation. I would object to regard an atom in a heart cell (let
alone a quark...) as something derived from that cell. I think that
this has much to do with ontological dependence: X derives from Y only
if Y is necessary for the existence of X:
- My heart derives from my body because it only exists in virtue of my body.
- The engine does not derive from my car because it already existed in
its entirety before the car came into existence

Kind regards,

Stefan

--
Stefan SCHULZ (apl. Prof. Dr. med.)
Institute of Medical Biometry and Medical Informatics
University Medical Center Freiburg
Stefan-Meier-Strasse 26 79104 Freiburg (Germany)
[home: Eschholzstr. 70, D-79115 Freiburg]
+49 (0)761 2036725, 2049089
http://purl.org/steschu
[stsc...@uni-freiburg.de], Skype: stschulz

------------------------------------------------------------------------------
Download Intel&#174; Parallel Studio Eval
Try the new software tools for yourself. Speed compiling, find bugs
proactively, and fine-tune applications for parallel performance.
See why Intel Parallel Studio got high marks during beta.
http://p.sf.net/sfu/intel-sw-dev
_______________________________________________
Obo-taxonomy mailing list
Obo-ta...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/obo-taxonomy

[Arlin Stoltzfus]

On Mar 15, 2010, at 6:00 AM, Stefan Schulz wrote:

> Second, a technical one: DL classifiers cannot deal with metaclasses,
> and it seems that nobody wants to open Pandora's box and allow "sets
> of sets" (Russel's paradox)

This is going to make my brain hurt even more, but ... I thought that
Russell's (2 L's) paradox only applied to an older version of set
theory before the modern axiomatizations that eliminated the
possibility of Russell's paradox. Does it really apply to DL
classifications?

> But I think, that the whole discussion boils down to the notion of
> derivation. I would object to regard an atom in a heart cell (let
> alone a quark...) as something derived from that cell. I think that
> this has much to do with ontological dependence: X derives from Y only
> if Y is necessary for the existence of X:
> - My heart derives from my body because it only exists in virtue of
> my body.
> - The engine does not derive from my car because it already existed in
> its entirety before the car came into existence

I'm confused. Apparently you are not the recipient of a heart
transplant (congratulations), but you might be a donor someday, in
which case your heart will exist outside of your body. What does
"exists in virtue of" have to do with "derives_from", which is a
relationship across a temporal divide?

Arlin
-------
Arlin Stoltzfus (stol...@umbi.umd.edu)
Fellow, CARB; Adj. Assoc. Prof., UMBI; Research Biologist, NIST
CARB, 9600 Gudelsky Drive, Rockville, MD
tel: 240 314 6208; web: www.molevol.org

[Chris Mungall]

On Mar 15, 2010, at 1:24 PM, Arlin Stoltzfus wrote:

> On Mar 15, 2010, at 6:00 AM, Stefan Schulz wrote:
>
>> Second, a technical one: DL classifiers cannot deal with metaclasses,
>> and it seems that nobody wants to open Pandora's box and allow "sets
>> of sets" (Russel's paradox)
>
> This is going to make my brain hurt even more, but ... I thought that
> Russell's (2 L's) paradox only applied to an older version of set
> theory before the modern axiomatizations that eliminated the
> possibility of Russell's paradox. Does it really apply to DL
> classifications?

OWL2-DL allows us to treat classes as instances via 'punning'. I
believe this is powerful enough to allow for the required amount of
metaclass reasoning without introducing Russel's paradox.

But this may be moot depending on exactly what kinds of reasoning one
wanted to do at the metaclass level. One the one hand we have very
simple use cases such as being able to retrieve all classes that are
instances of GenusRankedClass. On the other hand we might want to be
able to infer which taxon classes are instances of
HomophyleticTaxonClass, which (I believe) is outside what can be done
in OWL2, requiring some extension like description graphs.

>> But I think, that the whole discussion boils down to the notion of
>> derivation. I would object to regard an atom in a heart cell (let
>> alone a quark...) as something derived from that cell. I think that
>> this has much to do with ontological dependence: X derives from Y
>> only
>> if Y is necessary for the existence of X:
>> - My heart derives from my body because it only exists in virtue of
>> my body.
>> - The engine does not derive from my car because it already existed
>> in
>> its entirety before the car came into existence
>
>
> I'm confused. Apparently you are not the recipient of a heart
> transplant (congratulations), but you might be a donor someday, in
> which case your heart will exist outside of your body. What does
> "exists in virtue of" have to do with "derives_from", which is a
> relationship across a temporal divide?

I too am confused here..

[Stefan Schulz]

Arlin,

>> But I think, that the whole discussion boils down to the notion of
>> derivation. I would object to regard an atom in a heart cell (let
>> alone a quark...) as something derived from that cell. I think that
>> this has much to do with ontological dependence: X derives from Y only
>> if Y is necessary for the existence of X:
>> - My heart derives from my body because it only exists in virtue of
>> my body.
>> - The engine does not derive from my car because it already existed in
>> its entirety before the car came into existence
>
> I'm confused.  Apparently you are not the recipient of a heart
> transplant (congratulations), but you might be a donor someday, in
> which case your heart will exist outside of your body.

In the case I have a heart transplant, the heart is part of my body,
but it is not derived from my body, just as the engine of the car is
part of it but not derived from it.
No idea how you infer that the heart exists outside of my body.

> What does
> "exists in virtue of" have to do with "derives_from", which is a
> relationship across a temporal divide?

You are right, derives-from (as meant in OBO-RO) is not the correct
relation here.
The relation we need here is something like "comes-into-being-as-part-of"

Stefan

[Alan Ruttenberg]

On Mon, Mar 15, 2010 at 5:04 PM, Chris Mungall <c...@berkeleybop.org> wrote:
>
> On Mar 15, 2010, at 1:24 PM, Arlin Stoltzfus wrote:
>
>> On Mar 15, 2010, at 6:00 AM, Stefan Schulz wrote:
>>
>>> Second, a technical one: DL classifiers cannot deal with metaclasses,
>>> and it seems that nobody wants to open Pandora's box and allow "sets
>>> of sets" (Russel's paradox)
>>
>> This is going to make my brain hurt even more, but ... I thought that
>> Russell's (2 L's) paradox only applied to an older version of set
>> theory before the modern axiomatizations that eliminated the
>> possibility of Russell's paradox.  Does it really apply to DL
>> classifications?
>
> OWL2-DL allows us to treat classes as instances via 'punning'. I
> believe this is powerful enough to allow for the required amount of
> metaclass reasoning without introducing Russel's paradox.

It lets you have individuals that have the same name as classes. But
those individuals are not related to classes - no inference about the
class with the same name of an individual is made based on anything
said about the individual.

There are various features that make it *look* like that. For example,
annotations are associated with names, rather than individual,
classes, or properties. An since classes and individuals can share
names, it looks like they share annotations.

So really, there is no metaclass-like reasoning. However by not
forcing the name of classes and instances to be disjoint OWL-2 reduces
the barrier to outside-of-DL reasoning and query that lets people do
some of what they want.

> But this may be moot depending on exactly what kinds of reasoning one
> wanted to do at the metaclass level. One the one hand we have very
> simple use cases such as being able to retrieve all classes that are
> instances of GenusRankedClass.

Again, sort of. This assumes that you rigorously, outside of the
logic, remember to use the name for both class and instance. But it is
perfectly possible, for example, to have instances of
GenusRankedClass whose names do not in fact name any owl:Class.

> On the other hand we might want to be
> able to infer which taxon classes are instances of
> HomophyleticTaxonClass, which (I believe) is outside what can be done
> in OWL2, requiring some extension like description graphs.

I think that's right. I have a vague memory that it's possible to
define classes to be monophyletic, but I don't think you can define a
class which includes all instances of such classes.

-Alan