[Obo-discuss] Ontological Realism and OBO Foundry Criteria
Robert Hoehndorf
The OBO Foundry has, as a community, decided to embrace a set of
criteria that must be satisfied by all of its members. Some criteria
have been discussed for several years. In particular the criteria
pertaining to one form of "realism" [1] have been subject to
considerable debate.
Recently, the form of realism embraced by the OBO Foundry has been
subject to criticisms [2,3], which argue that these criteria are
neither necessary nor desirable guidelines for the construction of
scientific ontologies, but are often harmful for ontologies in
science.
In light of this new scientific evidence, do you think that the
criteria could be changed?
Michel and Rob.
[1] http://obofoundry.org/wiki/index.php/OBO_Foundry_Principles
criteria 6 and 10 from First OBO Foundry Summit, 2008
[2] http://iospress.metapress.com/content/j3324564p5l33863/
[3] http://dumontierlab.com/pdf/2010_FOIS_realism.pdf
------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
_______________________________________________
Obo-discuss mailing list
Obo-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/obo-discuss
Chris Mungall
The principle on the wiki page is:
Terms in an ontology should correspond to instances in reality
This doesn't sound especially controversial to me.
I'm afraid I only skimmed the papers below. Perhaps for those of us
without the philosophical background you can distill this a little
further. Can you give some concrete examples of how adhering to this
principle would lead to problems for practical users and developers of
bio-ontologies? For example, could the gene ontology be improved by
adding terms that do not correspond to instances in reality?
Michel Dumontier
Nicolas Le novère
In the past, I have been a criticism of the realist position, being rather
a rationalist myself, and developing ontologies which realm is the human
mind (SBO, KiSAO, TEDDY ...) However, I believe the OBO principle
"Terms in an ontology should correspond to instances in reality"
is here analysed in a narrower sense than intended (OBO realists are of
course welcome to disagree and correct me). A term describing a chemical
compound that "could" have instances in the reality is OBO-compliant. Even
if it has never been synthetised by anyone. What is not OBO-compliant is a
term that cannot have instances in reality. The classical example is the
"absent hand". The "absent hand" is not an "hand". Because if it is absent,
it does not exist. The "absent hand" is not an "abnormal hand". The "arm
with not hand" is an "abnormal arm". On the contrary, the "phantom hand" of
an amputee is a proper OBO term. Because despite there is no physical
structure located at the spatial coordinates of the former hand, the
feeling of the "phantom hand" is real in the brain of the amputee (we can
actually record it with fMRI). Of course the "phantom hand" is not
"abnormal hand". It is a "sensation".
IMHO, Michel, your example is perfectly valid in OBO (please OBO geeks, say
it is, otherwise we indeed have a huge problem).
> Visit sprint.com/first <http://sprint.com/first> --
> http://p.sf.net/sfu/sprint-com-first
> _______________________________________________
> Obo-discuss mailing list
> Obo-d...@lists.sourceforge.net
> <mailto:Obo-d...@lists.sourceforge.net>
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
>
>
>
>
>
> --
> Michel Dumontier
> Associate Professor of Bioinformatics
> Carleton University
> http://dumontierlab.com
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
>
>
>
> _______________________________________________
> Obo-discuss mailing list
> Obo-d...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
Colin Batchelor
> IMHO, Michel, your example is perfectly valid in OBO (please OBO geeks,
> say
> it is, otherwise we indeed have a huge problem).
Michel seems to be claiming that molfiles, predictions, hypotheses and so forth don't exist in reality.
That can't be right. Sober and thoroughgoing realists often write about promises, conventions, laws, contracts, beliefs, none of which are any less real than molfiles, predictions, hypotheses and so forth.
Colin.
--
Dr Colin Batchelor MChem CChem CSci,
Senior Informatics Analyst,
Royal Society of Chemistry, Thomas Graham House, Cambridge, UK CB4 0WF.
t: +44 1223 432280 e: batch...@rsc.org
DISCLAIMER:
This communication (including any attachments) is intended for the use of the addressee only and may contain confidential, privileged or copyright material. It may not be relied upon or disclosed to any other person without the consent of the RSC. If you have received it in error, please contact us immediately. Any advice given by the RSC has been carefully formulated but is necessarily based on the information available, and the RSC cannot be held responsible for accuracy or completeness. In this respect, the RSC owes no duty of care and shall not be liable for any resulting damage or loss. The RSC acknowledges that a disclaimer cannot restrict liability at law for personal injury or death arising through a finding of negligence. The RSC does not warrant that its emails or attachments are Virus-free: Please rely on your own screening.
Peter Robinson
> Nicolas Le novère writes:
>
>> IMHO, Michel, your example is perfectly valid in OBO (please OBO geeks,
>> say
>> it is, otherwise we indeed have a huge problem).
>
> Michel seems to be claiming that molfiles, predictions, hypotheses and so forth don't exist in reality.
>
> That can't be right. Sober and thoroughgoing realists often write about promises, conventions, laws, contracts, beliefs, none of which are any less real than molfiles, predictions, hypotheses and so forth.
>
Are things that are absent then the only things that do not exist in an
ontological sense?
:-0
-peter
> Colin.
>
--
PD Dr. med. Peter N. Robinson, MSc.
Institut für Medizinische Genetik
Charité - Universitätsmedizin Berlin
Augustenburger Platz 1
13353 Berlin
Germany
voice: 49-30-450566042
fax: 49-30-450569915
email: peter.r...@charite.de
http://compbio.charite.de/
http://www.human-phenotype-ontology.org
Robert Hoehndorf
Hi,
> > IMHO, Michel, your example is perfectly valid in OBO (please OBO geeks,
> > say
> > it is, otherwise we indeed have a huge problem).
> Michel seems to be claiming that molfiles, predictions, hypotheses and so forth don't exist in reality.
> That can't be right. Sober and thoroughgoing realists often write about promises, conventions, laws, contracts, beliefs, none of which are any less real than molfiles, predictions, hypotheses and so forth.
Quite right. And if we are going to build ontologies to support
science by providing a formal characterization of the meaning of the
terms used in scientific discourse, we need not only describe the
features of the hypotheses themselves (e.g., that they exist in minds,
that some consist of 5 axioms, that one particular hypothesis was
conceived by Darwin, etc.), but also their contents, i.e., what they
are about. But predicted classes of not yet synthesized chemicals
must not be present in a "realism-based" ontology [1], and therefore
you cannot define a hypothesis about this chemical, either.
This problem is one of the fundamental problems of the IAO, which
fails -- at least partially because of this OBO criterion, I would
conjecture -- to define its classes properly. For an example, look at
"numeral" and its subclass "integer numeral", which are defined, in
natural language, as being "about" numbers and integer numbers,
respectively (and even NLP could infer the is-a relation between
integer and number from the text alone). But the formal definition is
missing, presumably because "number" has no instance in reality. And
even if numbers were included in some ontology because they are found
to be "real", IAO would never be applicable for representing
hypotheses about things which do not (yet) exist, only because of the
criterion 6 (unless, of course, IAO is satisfied with purely textual
definitions, because apparently it seems okay to refer to non-real
things in natural language).
If, on the other hand, "corresponds to instances in reality" should
mean "could have instances in reality", as Nicolas suggests, I think
it should be reflected in the formulation of the criterion to avoid
further confusion (and the 10th criterion, "ontologies consist of
representations of types in reality" could be rephrased as well).
Rob.
[1] Quote from: Barry Smith, Waclaw Kusnierczyk, Daniel Schober, and
Werner Ceusters. Towards a reference terminology for ontology research
and development in the biomedical domain. In Proceedings of KR-MED
2006, 2006.
"Designer drugs are conceived, modeled, and described long before
they are successfully synthesized, and the plans of pharmaceutical
companies may contain putative references to the corresponding
chemical universals long before there are instances in reality. But
again: such descriptions and plans can be perfectly well apprehended
even within terminologies and ontologies conceived as relating
exclusively to what is real. Descriptions and plans do, after all,
exist. On the other hand it would be an error to include in a
scientific ontology of drugs terms referring to pharmaceutical
products which do not yet (and may never) exist, solely on the basis
of plans and descriptions. Rather, such terms should be included
precisely at the point where the corresponding instances do indeed
exist in reality, exactly in accordance with our proposals above."
James Malone
really do not care. Furthermore, that if approaches constrain ontology
development to the extent that a user can not have his or her classes
added to an ontology then they will be unhappy (and will often develop
something else instead). Moreover, if an approach necessitates that, in
order to be compatible with a framework, a formal representation becomes
so complex that it is no longer recognisable as the entity it is
intended to represent, then it is not useful to the community.
My personal opinion on realism is mostly irrelevant but my primary
concern is that users of ontologies I develop have their use cases
satisfied. If Michel's example is indeed incompatible with Realism-type
approaches then I think this is a problem as this is a valid use case
for the biomedical community. I have seen discussion on others such as
the null hypothesis.
So moving forward, I would propose the following; a set of legitimate
use cases from the community should be collected which are proposed as
not satisfiable by Realism type criterion and evaluated. This seems to
me an objective approach to evaluating whether or not a framework is
compatible, BFO or otherwise. This might help to remove discussions
about philosophy which tend to engender strong personal opinion which
make objective evaluation very difficult.
Cheers,
James
Peter Robinson wrote:
> On 07/15/2010 05:09 PM, Colin Batchelor wrote:
>
>> Nicolas Le novère writes:
>>
>>
>>> IMHO, Michel, your example is perfectly valid in OBO (please OBO geeks,
>>> say
>>> it is, otherwise we indeed have a huge problem).
>>>
>> Michel seems to be claiming that molfiles, predictions, hypotheses and so forth don't exist in reality.
>>
>> That can't be right. Sober and thoroughgoing realists often write about promises, conventions, laws, contracts, beliefs, none of which are any less real than molfiles, predictions, hypotheses and so forth.
>>
>>
>
>
> Are things that are absent then the only things that do not exist in an
> ontological sense?
> :-0
> -peter
>
>
>
>
>
>
>
>> Colin.
>>
>>
>
>
>
--
European Bioinformatics Institute,
Wellcome Trust Genome Campus,
Cambridge, CB10 1SD,
United Kingdom
Tel: + 44 (0) 1223 494 676
Fax: + 44 (0) 1223 494 468
Michel_Dumontier
Indeed, were it sufficient that terms could be admitted which were in fact consistent with current scientific thought, and did not require having instances, I think this would go some ways to alleviate our immediate dilemma.
But let me pick up on your comment about things that do not exist, and argue that there is a need for such terms in our scientific ontologies. The example we gave in our FOIS presentation [1] had to do with published work about smallest saturated acyclic alkane that *can not* be made. These form important boundaries for experimentalists, but more importantly, uniformly guide our scientific theories. Again, not admitting such terms in an ontology does a disservice to science, which precludes effective communication.
[1] http://www.slideshare.net/micheldumontier/realism-for-scientific-ontologies
m.
Nicolas Le novère
Your pragmatic approach is attractive. Nevertheless, one should not forget
that ontologies are not just vocabularies or "classifications" but also
tools to reason. Therefore all requests by users are not to be honoured
without a bit of reflection. Been there, done that. Four years ago, I
incorporated a term in SBO because it was requested by a colleague, who was
also very influential (and became even more so since). This opened a
Pandora box that we are still struggling to close and which is plaguing the
proper use of the ontology.
I am very much in favour of taking an extended view of reality (*), and
this is maybe where the solution rests. In SBO, we have terms like
"Michaelis-Menten rate-law". And certainly, if I use the Michaelis-Menten
equation in a manuscript, with certain values for Kmax and Vmax, this is an
instance of what is described by the SBO term, that exists in "reality" (on
the paper), even if what exists in the "biological reality" is an enzymatic
reaction with a certain velocity.
But if one is using anatomical and physiological ontologies to integrate
various datasets and reason on them, the use of things like absent hands
may result in very funny situations with concentration of chemicals in the
blood of the absent hand.
(*) I personally do not believe in Aristotle universals. There is nothing
that we can describe that is independent of our minds. Everything is a
shared agreement between the set of electromagnetic activity in my mind in
response to a sensory perception, and the equivalent in your mind. There is
certainly something rather than nothing. But we use ontologies to describe
the impact this something has on our cerebral activity.
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Robert Hoehndorf
Hi Peter,
> Are things that are absent then the only things that do not exist in an
> ontological sense?
> :-0
In my opinion, "absent arm" is not an ontological problem at all. It
is a problem of semantics and formal definitions. We create ontologies
to specify the meaning of terms in a vocabulary, and "absent arm" is
used in medical terminology, and therefore it has a place in an
ontology, I think.
We usually use a variant of first order logics to formally represent
our ontologies, and in FOL, our predicates reach over individuals in a
world structure (which is any non-empty set). When introducing "absent
arm", the naive approach is to introduce some kind of "absent arm"
predicate. In FOL, it would be AbsentArm(x), i.e., a predicate that
applies to individuals in the world structure (unless we treat natural
language statements like "absent arm" as propositions, i.e., "there
does not exist an arm", and define AbsentArm(x) accordingly).
Yet this does not coincide with our intentions, because "absent arm"
is /used/ by scientists to refer to a state where there is no arm (and
no entity to refer to).
This does not mean that "absent arm" is something that cannot be used
and introduced in an ontology, and it is only here that ontological
analyses come into play, i.e., to analyze and specify what a
scientists /means/ when she uses "absent arm". I think it would be
along the lines of "not having an arm as part" or "Quality of things
that have no arm as part", but this is a debate for another time.
Rob.
Colin Batchelor
> but also their contents, i.e., what they are about. But predicted classes > of not yet synthesized chemicals must not be present in a "realism-based"
> ontology [1], and therefore you cannot define a hypothesis about this
> chemical, either.
Let us take a binding site on a protein. That exists. Equally the disposition to bind to that site exists. All sorts of things will have that disposition; molecules are sticky.
So (molecule and has_disposition (some to_bind_to_X_binding_site)) is on the side of reality.
So does the connection table for your non-existent molecule. The hypothesis is rather that the effective specification worked out by a real chemist based on the connection table will make a (molecule and has_disposition (some to_bind_to_X_binding_site)).
There. Everything is on the side of reality.
(A bit of refinement is needed to cope with selective binding, but that's the story in outline.)
Remember also that the synthetic plan the chemist comes up with may well make something that doesn't match the original connection table for all sorts of reasons. The whole process is rather more subtle than the cartoon presented by docking software.
Best wishes,
Colin.
DISCLAIMER:
This communication (including any attachments) is intended for the use of the addressee only and may contain confidential, privileged or copyright material. It may not be relied upon or disclosed to any other person without the consent of the RSC. If you have received it in error, please contact us immediately. Any advice given by the RSC has been carefully formulated but is necessarily based on the information available, and the RSC cannot be held responsible for accuracy or completeness. In this respect, the RSC owes no duty of care and shall not be liable for any resulting damage or loss. The RSC acknowledges that a disclaimer cannot restrict liability at law for personal injury or death arising through a finding of negligence. The RSC does not warrant that its emails or attachments are Virus-free: Please rely on your own screening.
------------------------------------------------------------------------------
Robert Hoehndorf
Hi,
> But if one is using anatomical and physiological ontologies to integrate
> various datasets and reason on them, the use of things like absent hands
> may result in very funny situations with concentration of chemicals in the
> blood of the absent hand.
But this is not a problem of ontology. It would just be false
formalization of a class, a mistake, something that should be fixed so
that the term "absent hand" corresponds to the intented meaning (that
there is no hand).
> (*) I personally do not believe in Aristotle universals. There is nothing
And such a believe should not be necessary to do science or to build
ontologies. And neither does it make you an anti-realist that you do
not believe in universals. The Applied Ontology paper [1] by Gary
Merrill nicely shows how two different meanings of "realism" have been
entangled over the last years in the form of ontological realism:
realism about Aristotelian universals (universals exist independent
from our minds) and realism about the empirical world (something
exists independent from our mind). It is quite possible to be a
realist "about the world" without embracing entities like
universals. And, as [1] shows, while belief in the empirical world is
indeed necessary to do science, belief in universals is quite
unnecessary (and belongs to the realm of philosophical, not scientific
debate, and may better be left on a meta-level to science as an
explanation of what science is doing).
Rob.
[1] http://iospress.metapress.com/content/j3324564p5l33863
Robert Hoehndorf
Hi,
> Let us take a binding site on a protein. That exists. Equally the disposition to bind to that site exists. All sorts of things will have that disposition; molecules are sticky.
> So (molecule and has_disposition (some to_bind_to_X_binding_site)) is on the side of reality.
> So does the connection table for your non-existent molecule. The hypothesis is rather that the effective specification worked out by a real chemist based on the connection table will make a (molecule and has_disposition (some to_bind_to_X_binding_site)).
> There. Everything is on the side of reality.
Only if you are a) not interested in designing something that has new
capabilities, b) are not interested in the structure of the molecule
you are designing, and neither any other chemical properties, and c)
you are not even interested in /what/ you are specifying with your
connection table.
Even simple things like "weight" are properties of a molecule, and not
of a connection table and not of a process plan in the mind of the
chemist.
Rob.
Chris Mungall
I like this suggestion a lot. Look forward to seeing the list of use
cases!
Cheers
Chris
Chris Mungall
On Jul 15, 2010, at 7:53 AM, Nicolas Le novère wrote:
> Hello,
>
> In the past, I have been a criticism of the realist position, being
> rather
> a rationalist myself, and developing ontologies which realm is the
> human
> mind (SBO, KiSAO, TEDDY ...) However, I believe the OBO principle
>
> "Terms in an ontology should correspond to instances in reality"
>
> is here analysed in a narrower sense than intended (OBO realists are
> of
> course welcome to disagree and correct me). A term describing a
> chemical
> compound that "could" have instances in the reality is OBO-
> compliant. Even
> if it has never been synthetised by anyone.
This seems fine to me. It is good practice to record the evidence for
the existence of a class, so that experimentally characterized
compounds can be distinguished from predicted ones.
If it's later discovered that the predicted compound would have
properties that are physically impossible to instantiate then the
class would be retired from the ontology. You could still keep a
record of it of course, and even lift it up to the level of a refuted
hypothetical compound, which exists in reality - as some kind of
information artefact.
> What is not OBO-compliant is a
> term that cannot have instances in reality. The classical example is
> the
> "absent hand". The "absent hand" is not an "hand". Because if it is
> absent,
> it does not exist. The "absent hand" is not an "abnormal hand". The
> "arm
> with not hand" is an "abnormal arm". On the contrary, the "phantom
> hand" of
> an amputee is a proper OBO term. Because despite there is no physical
> structure located at the spatial coordinates of the former hand, the
> feeling of the "phantom hand" is real in the brain of the amputee
> (we can
> actually record it with fMRI). Of course the "phantom hand" is not
> "abnormal hand". It is a "sensation".
I would be even more liberal and allow the string "absent hand" as a
label so long as it was defined clearly, something like the condition
of having zero hands on the end of a limb.
I would object to something like:
an absent hand is a hand that is not there
But more for the boring reason that this leads to inference errors,
rather than because I'm defending some piece of metaphysical turf.
Perhaps it's the case that my personal interpretation of the realism
criteria is weak enough to render it superfluous and harmless. To me
it's a useful way of avoiding confusions between concepts and the
things those concepts are about, and also a way to properly treat
things like "gender unknown". There may be other ways of doing this,
insisting on an audit trail to reality is (I thought) simple and
uncontroversial.
Barry Smith
Dumontier/Hoehndorf at the FOIS meeting, and also
to the parallel arguments recently put forward by
Gary Merril, is being written up as we speak, and
should be ready in the next few days. The cases
so far presented are easy to solve, and do not
have anything specifically to do with realism, or
with universals, or with general terms, but
rather with general problems involved in
referring to non-existents (including those
non-existents which will, or might, exist in the
future, which we often need to talk about e.g.
when planning, e.g. in chemistry).
Watch this space.
BS
Nicolas Le novère
> But let me pick up on your comment about things that do not exist, and
> argue that there is a need for such terms in our scientific ontologies.
> The example we gave in our FOIS presentation [1] had to do with
> published work about " that *can not* be
> made. These form important boundaries for experimentalists, but more
> importantly, uniformly guide our scientific theories. Again, not
> admitting such terms in an ontology does a disservice to science, which
> precludes effective communication.
Hi Michel,
But the "smallest saturated acyclic alkane" that can not be made is a valid
subclass of the "chemical compound that cannot be made. I would still
consider it part of reality if we can draw them on paper, even if they
cannot be synthetized.
I would still be uncomfortable with it to be a subclass of "saturated
acyclic alkane". And here is the core of my position actually. Most of
those discussions could be solved by a careful positioning of the terms.
More later with the missing hand.
Nicolas Le novère
> In my opinion, "absent arm" is not an ontological problem at all. It
> is a problem of semantics and formal definitions. We create ontologies
> to specify the meaning of terms in a vocabulary, and "absent arm" is
> used in medical terminology, and therefore it has a place in an
> ontology, I think.
> Yet this does not coincide with our intentions, because "absent arm"
> is /used/ by scientists to refer to a state where there is no arm (and
> no entity to refer to).
That is not a good reason to adopt a term, for two reasons. The first is
that there is a large body of "knowledge" used by people, including by
"scientist" that is actually wrong. If you look at 10 textbook on
neurobiology, 9 will mentioned that the action potential is generated at
the level of the axon hillock (and this is what I leant in high school, and
again in university). Nevertheless this is wrong. Any specialist will tell
you that the action potential is initiated in the axon initial segment, and
not the axon hillock. But the important information here is that each time
I explained to someone what was the axon hillock, the initial segment and
why was the axon potential initiated at the level of the second, this
explanation was well received and the person changed her internal
representation. My point here is that scientists are eager to learn and
improve. And that brings me to the second point which is that one of the
missions of ontologies in biology is to help people thinking clearly. And
that works. Over the last decade, OBO helped me A LOT in properly
formulating my understanding of the world, and this influenced my research.
In summary I really disagree with the need of incorporating classes just
because "people" use them. But designing clean ontologies can help people a
lot, and they will gladly accept that.
> This does not mean that "absent arm" is something that cannot be used
> and introduced in an ontology, and it is only here that ontological
> analyses come into play, i.e., to analyze and specify what a
> scientists /means/ when she uses "absent arm". I think it would be
> along the lines of "not having an arm as part" or "Quality of things
> that have no arm as part", but this is a debate for another time.
Yes (for the last part) and no (for the first). As a parenthesis, let-me
say that we should distinguish between the ontology and the way it is
exposed to the end-user. Reversing your reasoning, we could generate an
"absent arm" at the level of the user display based on the "not having an
arm as part".
The problem here is the Russel element/set problem. You have elements and
sets. And they belong to different realms. An absent part does not exist.
However, a set without a part exists. If we have a class M, with three
classes as parts N1, N2, N3. m1, made up of n1, n2, n3 is an instance of M.
But m2, made of n1, n2 is also an instance of M. In other words an upper
limb with arm, forearm and a hand is an instance of an upper limb. An upper
limb with arm, forearm but no hand is still an instance of a arm. Now, the
interesting thing is that I believe an upper limb with no arm, no forearm
and no hand is still an instance of an upper limb. See? There is no "absent
upper limb". But there could be an upper limb with all parts that are
missing. I realise this is a trick. But I believe this could solve some
problems in a pragmatic way.
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Michel_Dumontier
I am reminded of Magritte's drawing titled "ceci n'est pas une pipe" - While making a simplified representation of molecular structure by drawing it out on paper is particularly informative to chemists, it is not a molecule nor is it the chemical structure. As it stands now, we are at a loss to describe that the drawing or representation *is about* a certain kind of chemical, or how the cartoon is a *specification* for the structure. Even though we can compute all sorts of qualities and attributes from the chemical graph, we cannot associate this with a type in a chemical ontology.
If people want to produce ontologies that conform to a highly constraining and fallible realist point of view, by somehow only working with "positive" terms or "repositioning" terms, then it's obviously up to them. But often this leads to an unnatural workaround for a philosophical position, and we don't believe it to be a particularly useful criteria for scientific ontologies nor in their applications. It is much more effective to formally describe what terms correspond to (whether they have instances or not).
Cheers,
m.
-----Original Message-----
From: Nicolas Le novère [mailto:le...@ebi.ac.uk]
Sent: Thursday, July 15, 2010 5:30 PM
To: obo-d...@lists.sourceforge.net
Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry Criteria
Chris Mungall
If your unicorns and impossible chemical structures have to live
outside the OBO Foundry is that such a bad thing for you? No one is
saying that all ontologies have to be OBO Foundry compliant ontologies.
But let's just say that your boss is telling you that your ontology
must be OBO Foundry compliant. I'm still not seeing what the problem
is. You can make your drawings and/or write your axioms describing
your proposed structure. You can represent the drawing as a drawing
and/or reify those axioms, and make additional statements like "this
structure can't exist because of property P, but if it were to exist
it might have properties Q, R and S". Is this an unnatural workaround?
Wouldn't a more direct representation lead to inconsistencies when
reasoning? You could have a direct representation that you use for
some purposes (e.g. discovering and explaining those inconsistencies),
and then some kind of automated translation of that to the reified
form for the purposes of integrating with other ontologies.
Is this so bad?
Chris Mungall
that don't adhere to the realism criteria -- but these will go into a
special category, which will have its own special logo.
Adam M. Goldstein
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
On Jul 15, 2010, at 20:03, Chris Mungall <c...@berkeleybop.org> wrote:
>
> If your unicorns and impossible chemical structures have to live
> outside the OBO Foundry is that such a bad thing for you? No one is
> saying that all ontologies have to be OBO Foundry compliant ontologies.
>
> But let's just say that your boss is telling you that your ontology
> must be OBO Foundry compliant. I'm still not seeing what the problem
> is. You can make your drawings and/or write your axioms describing
> your proposed structure. You can represent the drawing as a drawing
> and/or reify those axioms, and make additional statements like "this
> structure can't exist because of property P, but if it were to exist
> it might have properties Q, R and S". Is this an unnatural workaround?
I think there's a problem here because "can't" and "if it were" are modal expressions, and so far as I know there aren't any implementations of modal logics in formal reasoning. Moreover I think possibilities and counterfactual statements would be anathema to the point of view of those saying they are realists here.
The case of the picture raises issues about propositional attitudes more generally. Someone can certainly believe there are unicorns, or draw a picture of one, or rather, what they would look like if they were real. Well,
Adam M. Goldstein
Anyhow, the point I was making was that there are plenty of cases in which we would want to represent a statement like "He believes there are unicorns," for instance, when representing states of mind of psychiatric patients. Or: a cognitive science experiment in which some people are expected to give incorrect answers to sone questions. So to represent the state space for the outcomes of this experiment, there will have to be "he wrongly believes..."
So, I don't know how realists want to deal with this. My impression is that realists of the sort under consideration here don't want to allow thoughts of unreal or impossible things.
I'm not saying that there is no way for the realist to incorporate this into his or her view, but I think that any view is too strong that eliminates thoughts or intended representations of impossible or fictional objects, or fails to reconstruct them.
I take it that the unicorn graphic signals the end of the discussion, but I hope my remarks have been useful or interesting nonetheless.
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
Michel_Dumontier
m.
-----Original Message-----
From: Chris Mungall [mailto:c...@berkeleybop.org]
Sent: Thursday, July 15, 2010 8:30 PM
To: obo-d...@lists.sourceforge.net
Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry Criteria
Nicolas Le novère
> If your unicorns and impossible chemical structures have to live
> outside the OBO Foundry is that such a bad thing for you? No one is
> saying that all ontologies have to be OBO Foundry compliant ontologies.
It is a bad thing actually. OBO Foundry, whether intended or not, has
become a standardisation body. And I believe everyone working on OBO
ontologies have the goal of developing a system of orthogonal ontologies,
where each of them is the reference in the field. And this is already the
case for ChEBI. All the chemicals in our models are annotated with ChEBI.
So if there is a model of a biological process dealing with a non-existent
chemical (impossible of not is not the issue here), we need this ChEBI
entry. Furthermore, the Innovative Medicine Initiative is launching a very
large project intended to develop an infrastructure allowing to follow
PK/PD models, from the initial idea to the clinical phase III. In this
project, the models and associated data are linked to industry standards
such as CDISK and HL7 through OBO foundry ontologies (or candidate
ontologies). We will therefore use ChEBI together with compounds that have
never been synthesised, and many will never be. But being able to identify
these compounds and relating them using ChEBI terms will be absolutely
crucial. Those are just two examples among plenty. And this does not
(should not) infringe the realist position at all. If the compound has been
designed in silico, and studied so much we can have QSAR information, it
really exists.
Nicolas Le novère
Janna Hastings
Hello all,
As regards the following quote Rob cites, I would like to point out that a
planned or predicted molecule is definitely not a drug product, and would
rightly be excluded from any ontology of drug products. Drug products are
those things that have brand names, packaging, dosage recommendations,
package inserts listing side effects and have been approved for various
indications by a legal body which has the power to endorse drugs.
I cannot imagine any pharmaceutical application use case for which an
ontology of *drug products* would benefit from the inclusion of millions
of not-yet-synthesised specifications from a combinatorial chemical
library.
Cheers, Janna
Janna Hastings
> that there is a large body of "knowledge" used by people, including by
> "scientist" that is actually wrong. If you look at 10 textbook on
> neurobiology, 9 will mentioned that the action potential is generated at
> the level of the axon hillock (and this is what I leant in high school,
> and
> again in university). Nevertheless this is wrong. Any specialist will tell
> you that the action potential is initiated in the axon initial segment,
> and
> not the axon hillock. But the important information here is that each time
> I explained to someone what was the axon hillock, the initial segment and
> why was the axon potential initiated at the level of the second, this
> explanation was well received and the person changed her internal
> representation. My point here is that scientists are eager to learn and
> improve. And that brings me to the second point which is that one of the
> missions of ontologies in biology is to help people thinking clearly. And
> that works. Over the last decade, OBO helped me A LOT in properly
> formulating my understanding of the world, and this influenced my
> research.
> In summary I really disagree with the need of incorporating classes just
> because "people" use them. But designing clean ontologies can help people
> a
> lot, and they will gladly accept that.
>
Very well put!
Janna
Matthias Samwald
> and/or reify those axioms, and make additional statements like "this
> structure can't exist because of property P, but if it were to exist
> it might have properties Q, R and S". Is this an unnatural workaround?
I guess the proponents of representing possibilia in OBO ontologies don't
like to 'represent the drawing as a drawing' because that would require an
entirely new ontology of 'molecules on paper'. The class 'molecule' would be
reflected in a new class 'molecule on paper', the property 'molecular
weight' would be reflected in a property 'molecular weight on paper of a
molecule on paper', or something to that effect. Indeed, that would be very
unnatural and silly.
The far more natural way to represent possibilia would be the second choice
you offer, reification. If we agree that reification is sanctioned by the
OBO Foundry to represent possibilia, then we should start to talk about good
practices for doing so. Does it mean that we actually CAN capture unicorns
in OBO ontologies as long as they are reified? What is the best way to
represent it in OWL? And is it even possible to represent reification in the
OBO format?
Cheers,
Matthias Samwald
// DERI Galway, Ireland
// Konrad Lorenz Institute for Evolution and Cognition Research, Austria
// http://samwald.info
Ravensara Travillian
Raven
--
Ravensara S. Travillian
Vertebrate Anatomist/Bioinformatician
European Bioinformatics Institute,
Wellcome Trust Genome Campus,
United Kingdom
Tel: + 44 (0) 1223 494 553
David Osumi-Sutherland
> Classy. You must feel proud of this achievement. Another tactical
> failure in addressing the issue at hand.
I'm glad that you and Robert have brought these issues up. There are a
number of issues in your paper and the Merrel paper that deserve good
answers. And the discussion so far has covered some of them
constructively. But I disagree with you here. I think Chris' post
does address one of the issues at hand.
I presume from your reaction that you wouldn't have a favorable
opinion of an ontology that contained assertions:
unicorn
. SubClassOf: has_part some rainbows
. SubClassOf: has_part some magic
. SubClassOf: has_part some superglue
...
This suggests you have some boundary criteria for ontologies, some
criteria judging what makes one ontology good and another one not.
As I see it, the OBO Foundry has a simple answer: exclude fiction.
What's yours?
To give a more every day example. The following are a few of the
classes obsoleted from the Drosophila anatomy ontology because
instances do not exist. Would you advise keeping them, and if not,
what would be your grounds for removing them?
[Term]
id: FBbt:00000373
name: lateral G branch 2
comment: According to Manning and Krasnow, 1993 (FBrf0064787; page
628), no lateral ganglionic branch exists in tracheal metamere 2.
subset: cur
synonym: "LG2" RELATED []
is_obsolete: true
consider: FBbt:00000372
[Term]
id: FBbt:00000374
name: lateral G branch 3
comment: According to Manning and Krasnow, 1993 (FBrf0064787; page
628), no lateral ganglionic branch exists in tracheal metamere 3.
subset: cur
synonym: "LG3" RELATED []
is_obsolete: true
consider: FBbt:00000372
[Term]
id: FBbt:00000414
name: non-functional larval spiracle of thoracic segment 2
comment: Does not exist.
subset: cur
is_obsolete: true
consider: FBbt:00000412
[Term]
id: FBbt:00002815
name: abdominal basiconical sensillum dbd
comment: This term was obsoleted because abdominal basiconical
sensillum dbd does not exist. Possible replacement terms are:
prothoracic basiconical sensillum dbd, the only segment with a dorsal
group basiconic sensillum is t1; abdominal dorsal bipolar neuron dbp,
which is referred to as dbd in some publications.
subset: cur
is_obsolete: true
consider: FBbt:00002442
consider: FBbt:00002728
[Term]
id: FBbt:00004820
name: adult abdominal spiracle 8
comment: There are only 7 pairs of spiracles in the adult abdomen.
This is true for both sexes. (See FBrf0007734, pg409.)
subset: cur
is_obsolete: true
consider: FBbt:00004814
[Term]
id: FBbt:00004821
name: adult abdominal spiracle 9
comment: There are only 7 pairs of spiracles in the adult abdomen.
This is true for both sexes. (See FBrf0007734, pg409.)
subset: cur
is_obsolete: true
consider: FBbt:00004814
Cheers,
David
On 16 Jul 2010, at 16/Jul/2010 08:36:27, Michel_Dumontier wrote:
>
> m.
>
> -----Original Message-----
> From: Chris Mungall [mailto:c...@berkeleybop.org]
> Sent: Thursday, July 15, 2010 8:30 PM
> To: obo-d...@lists.sourceforge.net
> Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry
> Criteria
>
> I'm attaching new OBO Foundry logo. We've decided to admit
> ontologies that don't adhere to the realism criteria -- but these
> will go into a special category, which will have its own special logo.
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
> _______________________________________________
> Obo-discuss mailing list
> Obo-d...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
David Osumi-Sutherland, PhD
Curator/ Ontologist
FlyBase / Virtual Fly Brain
Department of Genetics,
University of Cambridge,
Downing Street,
Cambridge, CB2 3EH, UK
Tel: +44 (0)1223 333 963
Fax: +44 (0)1223 766 732
Janna Hastings
without answering your more anatomy-specific question, in terms of the OBO
Foundry policies, I think it is not the case that an entire ontology would
be accepted or rejected as additional knowledge rendered specific terms
invalid or obsolete. Rather, the OBO Foundry expects participating
ontologies to have a policy in place whereby they make terms obsolete if
new scientific knowledge renders them invalid. I suppose that failure to
do this over a long period of time despite multiple requests by users
might result in Foundry exclusion of a previously accepted ontology -- but
this seems very unlikely.
One of the OBO Foundry objectives is that ontologies are works in
progress, continuously evolving as the relevant domain science progresses.
Cheers, Janna
David Osumi-Sutherland
<morphospace.png>------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first_______________________________________________
Obo-discuss mailing list
Obo-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/obo-discuss
James Malone
"surprises" component in the diagram is more a posterior element...
I find this discussion is useful and I'm grateful for the original
posting. The OBO Foundry, as Nicolas has rightly pointed out, is seen by
many as a standards body and as such should also be open to scrutiny -
just as the ontologies that sit under its wing should. I know Barry et
al. have published a large body of work describing problems with
ontologies before Realist approaches and to the extent that I think
things have improved dramatically over the last few years, I agree with
him. Whether or not that is because of BFO I honestly do not know and it
is a reasonable, scientific question to ask whether we gain or lose by
this requirement; we should attempt to answer this objectively I
believe. I am grateful to Barry et al for their contribution to the
field and I am grateful to Michel, Rob et al for adding some discourse;
we are better for it.
My impression has always been that the Foundry are not interested in
ontologies in which classes are 'theoretical' and have no known
instances in reality (at present). Therefore, the work Michel and Rob
describe is likely out of scope. However, just because it is out of
scope for OBO, if indeed that is the case, does not make this work any
less valuable to the community. It is possible that in the future some
of the most useful classes will have first started life in an ontology
such as the one Michel describes and as such, should not be discouraged.
Cheers,
James
--
European Bioinformatics Institute,
Wellcome Trust Genome Campus,
Cambridge, CB10 1SD,
United Kingdom
Tel: + 44 (0) 1223 494 676
Robert Hoehndorf
Hi,
> > But let me pick up on your comment about things that do not exist, and
> > argue that there is a need for such terms in our scientific ontologies.
> > The example we gave in our FOIS presentation [1] had to do with
> > published work about " that *can not* be
> > made. These form important boundaries for experimentalists, but more
> > importantly, uniformly guide our scientific theories. Again, not
> > admitting such terms in an ontology does a disservice to science, which
> > precludes effective communication.
> But the "smallest saturated acyclic alkane" that can not be made is a valid
> subclass of the "chemical compound that cannot be made. I would still
> consider it part of reality if we can draw them on paper, even if they
> cannot be synthetized.
> I would still be uncomfortable with it to be a subclass of "saturated
> acyclic alkane". And here is the core of my position actually. Most of
> those discussions could be solved by a careful positioning of the terms.
> More later with the missing hand.
Yet this seems at the heart of the argument: the smallest saturated
acyclic alkane that cannot be made is a kind of alkane. Otherwise the
statement (and a scientific paper about it) would not make sense
(because it would be the "smallest molecule that cannot be made that
cannot exist". It is all about what problem ontologies are supposed to
solve: will they be catalogues of things which are believed to exist
by a certain group of scientists (all following a single scientific
theory), or do ontologies solve a communication and integration
problem.
In the latter case, an ontology would specify a class "saturated
acyclic alkane" and provide the properties of those, so that
scientists will be able to recognize one if they see one, design
experiments about them, etc. Some subclasses of "saturated acyclic
alkane" may have no instances, and it is a scientific result to show
that, with respect to a particular scientific theory, some subclasses
cannot have instances. Yet this would not make them suddenly change
their supertype: the class still satisfies all the properties of
"saturated acyclic alkane", if we used an OWL reasoner, it would still
infer that it is a kind of "alkane", etc. The /meaning" of "saturated
acyclic alkane" did not suddenly change because of this scientific
result. And your queries using the ontology should still return that
result: it is a molecule with certain properties, a subclass of
"saturated acyclic alkane". And only once you know what kind of thing
you are talking about (the ontological part) can you retrieve
information about the type from your database or application, such as
experiments conducted to synthesize it, or even impossibility results.
Catalogues of types of things that exist do not satisfy the same need,
they cannot be used to describe and annotate experiments or scientific
research results and hypothesis, because a class "saturated acyclic
alkane" would have to be defined so as to -- a priori -- exclude
things like those alkanes that cannot exist.
Rob.
Robert Hoehndorf
> If your unicorns and impossible chemical structures have to live
> outside the OBO Foundry is that such a bad thing for you? No one is
> saying that all ontologies have to be OBO Foundry compliant ontologies.
So we already arrived at the last line of defense: (1) unicorns and
(2) "if you don't like what we do, why don't you get out".
It is, of course, your right, as one of the elected OBO Foundry
editors, to defend the philosophical views held by the members of the
OBO Foundry community. But please note that nobody was talking about
unicorns so far, as there seems to be no need to account for "unicorn"
in a scientific ontology, because "unicorn" is not a term used in
scientific discourse or databases (if you know of unicorn scientists,
please let me know, otherwise please explain the relevance to the
discussion).
And for (2), I would have hoped that there was more to it than mere
belief in a philosophical view and its role in constructing
ontologies that made the OBO Foundry members and candidates chose this
criterion.
> But let's just say that your boss is telling you that your ontology
> must be OBO Foundry compliant. I'm still not seeing what the problem
> is. You can make your drawings and/or write your axioms describing
> your proposed structure. You can represent the drawing as a drawing
> and/or reify those axioms, and make additional statements like "this
Please explain what you mean by "reify". A drawing should always be a
drawing, and a molecule a molecule. Defining the class of drawings
/of/ a particular kind of molecule still requires two classes: that of
a drawing, and that of the molecule.
> structure can't exist because of property P, but if it were to exist
> it might have properties Q, R and S". Is this an unnatural workaround?
Yes, it is unnatural, and it likely does not work either, without
creating a class of the things that may not exist in "reality".
> Wouldn't a more direct representation lead to inconsistencies when
> reasoning? You could have a direct representation that you use for
> some purposes (e.g. discovering and explaining those inconsistencies),
> and then some kind of automated translation of that to the reified
> form for the purposes of integrating with other ontologies.
> Is this so bad?
As I explained, neither of these approaches seems to work, without
actually creating the class somewhere in the ontology, whether named
or unnamed.
And it still does not answer the question why the OBO Foundry, as a
community of scientists, has decided to embrace a particular
philosophical view about ontology, a view that several scientists
outside the OBO Foundry feel highly sceptical about.
Rob.
Janna Hastings
Hello,
>
> Yet this seems at the heart of the argument: the smallest saturated
> acyclic alkane that cannot be made is a kind of alkane. Otherwise the
> statement (and a scientific paper about it) would not make sense
> (because it would be the "smallest molecule that cannot be made that
> cannot exist". It is all about what problem ontologies are supposed to
> solve: will they be catalogues of things which are believed to exist
> by a certain group of scientists (all following a single scientific
> theory), or do ontologies solve a communication and integration
> problem.
But, trivially, a "molecule which cannot be made" is not a molecule! It
does not have a weight or any other properties, it cannot be placed in a
bottle or on my hand, and it will never be a part of a drug or have any
kind of bioactivity... etc.
To solve the communication problem, you may well wish to describe
predicted and impossible "molecules", but they are just that: predicted or
impossible, and not molecules. Molecules are spatially extended things
made up of atoms and electrons.
Janna
Robert Hoehndorf
Hi,
> As I see it, the OBO Foundry has a simple answer: exclude fiction.
> What's yours?
This is not about fiction, but science. Science makes predictions from
theories; they should be captured by ontologies because it is
important to communicate about them. Science shows impossibility
results as for perpetual motion, molecules, etc.; they should be
captured by ontologies, so that we can query our databases for them,
retrieve information about them, etc.
And there needs to be no other criteria once one has found to be
unnecessarily constraining.
Rob.
Michel_Dumontier
We wish to specify the meaning of terms in an ontology. Just because a term refers to a type that has no instances does not imply that it is invalid in a chemical ontology. Thus C16/C17 are kinds of molecules, with the additional knowledge that we know that they cannot persist. Instances in the real world would be spatially extended, and have all sorts of molecule-related properties. The type, as specified by its chemical graph, also have expected properties regarding its molecular weight, etc. absolutely.
m.
-----Original Message-----
From: Janna Hastings [mailto:hast...@ebi.ac.uk]
Sent: Friday, July 16, 2010 8:14 AM
To: obo-d...@lists.sourceforge.net
Cc: le...@ebi.ac.uk; obo-d...@lists.sourceforge.net
Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry Criteria
Ravensara Travillian
David Osumi-Sutherland wrote:
> This suggests you have some boundary criteria for ontologies, some
> criteria judging what makes one ontology good and another one not.
> As I see it, the OBO Foundry has a simple answer: exclude fiction.
> What's yours?
I am not sure that the boundary between "fiction" and "as-yet-unproven hypothesis that later proves useful" is all that clearly-defined when trying to evaluate it in light of current knowledge. Certainly, some things can be ruled out as definitively impossible by violating the laws of physics, but short of that, how do we distinguish "fiction" from "not-yet-demonstrated" a priori?
To take an example from the recent past, Schrödinger, expressing the views of many contemporary physicists, famously said about quantum mechanics, "I don't like it and I'm sorry I ever had anything to do with it.". Quantum mechanics seemed like badly-written fiction until it began delivering experimental results that classical physics could not account for. We know that _in retrospect_, but at the time, how would you (or the OBO) have distinguished quantum mechanics from fiction? If the policy is not to accept an ontology until it is definitively proven, then there are some domains which probably never will have enough instances to meet that criterion, and thus will never get accepted. I suspect evolutionary biology is one, just because there are so many gaps in the fossil record.
So it seems to me that your answer above is indeed concise, but I don't understand the criteria under which it is implemented (the definition of fiction). This discussion, is, however, generating some insights for me, and I do appreciate the participation in it.
> To give a more every day example. The following are a few of the
> classes obsoleted from the Drosophila anatomy ontology because
> instances do not exist. Would you advise keeping them, and if not,
> what would be your grounds for removing them?
Foreshadowing what Janna responds to me later below, it looks to me like an example of the process she describes. If this is indeed how OBO operates, then clearly, the criterion is not being applied strictly literally (since your obsoleted entries got into OBO in the first place without instances), and it is therefore not such a show-stopping problem as it first appeared to me to be.
Janna Hastings wrote:
> without answering your more anatomy-specific question, in terms of the
> OBO Foundry policies, I think it is not the case that an entire ontology
> would be accepted or rejected as additional knowledge rendered specific
> terms invalid or obsolete. Rather, the OBO Foundry expects participating
> ontologies to have a policy in place whereby they make terms obsolete if
> new scientific knowledge renders them invalid. I suppose that failure to
> do this over a long period of time despite multiple requests by users
> might result in Foundry exclusion of a previously accepted ontology -- but
> this seems very unlikely.
I agree that the scenario you describe is very unlikely. I was thinking more of one where we, at a certain state of knowledge, model the previously-agreed upon (phenetics-based, for example) tree of the relationships between chordates and echinoderms, and we model an entity, the LCA of chordates and echinoderms, way up the tree somewhere at the protostome/deuterostome branch. My understanding was that, as we did not have an actual instance of this LCA, the ontology would be considered non-compliant, and not accepted to OBO.
Then, when molecular evidence shows that they are much closer than previously thought, we have a re-worked tree structure and a *different* non-instanced LCA with different characteristics much closer to both echinoderms and chordates in version 2.0, plus an obsolete version of the ontology 1.0 that has been shown to have an incorrect LCA. I thought these would be deal-killers for inclusion in OBO; I am happy to hear that they are not, because each of these ontologies--although imperfect--has a role to play in hypothesis testing and evaluation to advance the science.
> One of the OBO Foundry objectives is that ontologies are works in
> progress, continuously evolving as the relevant domain science
> progresses.
That is very reassuring. I think the fervor of this discussion, however, demonstrates that that objective does not come through in the way the principle is currently stated; perhaps the Foundry should consider rewording it to more closely reflect the actual practice.
David wrote:
> Your issue then becomes an engineering
> question: how can I build a sustainable ontology - one unlikely to
> rapidly become stale as scientific understanding changes. The answer
> is avoid including highly contentious assertions in your ontology in
> the first place.
I think that staying away from highly-contentious assertions is very hard to do in evo-bio, and I also think that ontologies have a role to play in integrating data to provide evidence for and against these assertions. So I don't think saying some assertions are out of scope for ontologies really matches ontologies to the way scientists work. Additionally, we have the same problem as with "fiction" above--by what criterion do we determine highly-contentious vs. not-so-highly-contentious? Are some aspects of evo-bio fair game for ontology modeling, while others are not? How would a practising scientist distinguish among them?
> There are certainly plenty of assertions about common ancestors are
> highly contentious, but also plenty of others that are not. For
> example, I don't think it's contentious to say that the last common
> ancestor of you and a chicken had a forelimb with three segments, the
> most proximal of which consisted of a, single bone, the next of a pair
> of parallel bones and the most distal of numerous short bones.
Agreed. But those are all inferences, as it happened over 300 million years ago, and we don't have the complete record. So even as non-controversial an entity as the human-chicken LCA would be ruled out by definition, if having instances is an absolute requirement.
However, the obsoleted FlyBase examples you provide are reassuring that that is not actually the case, so I am thinking it is more a problem of communication than of substance at this point.
> Is the realism criterion to be read literally?
Based on the answers I've gotten so far, I'd say it looks like no.
> If so, does that mean a binary assumption of how morphospace is to
> be carved up? Does it assume only the white circle and the black
> circle, and nothing else?
I think there is a communication issue making it look this way, but that it is not actually the case, if OBO implements its principles in the way Chris, Janna, and David describe.
> If that is the case, does that mean that any evo-devo ontology I
> develop with provisional assumptions about LCAs, to be revised in
> light of later evidence [thus, falling in the grey circle], is by
> definition, not eligible for inclusion in OBO?
Provisionally no, based on Chris', Janna's, and David's responses.
> If it is not to be read literally, then can I take Chris'
> description of accepting an ontology and then revising it in light
> of later knowledge as an accurate statement of official OBO policy?
Provisionally yes, based on above.
> If that is the case, then what is the policy regarding earlier
> accepted ontologies that--only in light of later knowledge--are
> shown not to be OBO-compliant, yet were accepted?
Apparently, not an issue, assuming I am correctly understanding the implementation of the policies as described above.
I trust if this trajectory of this discussion does not actually reflect the policy of how that principle is interpreted, someone with the authority to speak for the OBO will weigh in with a correction.
Cheers,
Raven
--
Ravensara Travillian, PhD
Vertebrate Anatomist/Bioinformatician
European Bioinformatics Institute,
Wellcome Trust Genome Campus,
Hinxton, Cambridgeshire, CB10 1SD,
United Kingdom
Robert Hoehndorf
Hi,
> > Yet this seems at the heart of the argument: the smallest saturated
> > acyclic alkane that cannot be made is a kind of alkane. Otherwise the
> > statement (and a scientific paper about it) would not make sense
> > (because it would be the "smallest molecule that cannot be made that
> > cannot exist". It is all about what problem ontologies are supposed to
> > solve: will they be catalogues of things which are believed to exist
> > by a certain group of scientists (all following a single scientific
> > theory), or do ontologies solve a communication and integration
> > problem.
> But, trivially, a "molecule which cannot be made" is not a molecule! It
> does not have a weight or any other properties, it cannot be placed in a
> bottle or on my hand, and it will never be a part of a drug or have any
> kind of bioactivity... etc.
Maybe I was not very clear in my explanation, but your are mixing
types and their instances. In ontology, we characterize kinds of
entities by specifying the features of their instances. "Molecule" is
a type, and it can have several sub-types, which share the features of
"Molecule" (and therefore stand in the is-a relation to
Molecule). Some of these sub-types may not have instances in
"reality", yet they are, "trivially", sub-types of "Molecule". The
statement that they have no instances can only have meaning if they
are sub-types of "Molecule", because, "trivially", predictions, specs,
drawing, etc. all do have instances.
> To solve the communication problem, you may well wish to describe
> predicted and impossible "molecules", but they are just that: predicted or
> impossible, and not molecules. Molecules are spatially extended things
> made up of atoms and electrons.
And that kind of thing that cannot be created is a spatially extended
thing made up of atoms and electrons -- it is a subtype of "molecule".
Rob.
Robert Hoehndorf
Hi,
> My impression has always been that the Foundry are not interested in
> ontologies in which classes are 'theoretical' and have no known
> instances in reality (at present). Therefore, the work Michel and Rob
> describe is likely out of scope. However, just because it is out of
I would argue that ontologies like OBI or IAO, which attempt to
provide support for what scientists do (collect data, create theories,
make predictions from these theories, verify them by experiment),
would then be out of the OBO Foundry's scope, too.
I am not sure, but I think it is quite possible that even the GO has
classes without instances in reality, or at least without any /known/
instances. Good candidates are leaf classes without any
annotations. Classes which are included because they were predicted,
conjectured, hypothesized, and scientists are starting just now to
conduct experiments to learn more about them (if they have
instances). Maybe somebody with a better knowledge of biology or the
GO than me can attempt to find one.
Rob.
Colin Batchelor
> Yet this seems at the heart of the argument: the smallest saturated
> acyclic alkane that cannot be made is a kind of alkane.
Alkanes burn, bulk portions of them act as anaesthetics when inhaled, they react with chlorine radicals to yield chloroalkanes.
If you admit non-existent alkanes as kinds of alkanes then you get nonsensical entailments such as non-existent alkanes being able to burn, to yield chloroalkanes when reacting with chlorine radicals, to anaesthetize people when inhaled.
Now, if you'll excuse me I'm off to discuss whether he should have highlights with the present king of France.
best wishes,
Colin.
DISCLAIMER:
This communication (including any attachments) is intended for the use of the addressee only and may contain confidential, privileged or copyright material. It may not be relied upon or disclosed to any other person without the consent of the RSC. If you have received it in error, please contact us immediately. Any advice given by the RSC has been carefully formulated but is necessarily based on the information available, and the RSC cannot be held responsible for accuracy or completeness. In this respect, the RSC owes no duty of care and shall not be liable for any resulting damage or loss. The RSC acknowledges that a disclaimer cannot restrict liability at law for personal injury or death arising through a finding of negligence. The RSC does not warrant that its emails or attachments are Virus-free: Please rely on your own screening.
Robert Hoehndorf
Hi,
> > Yet this seems at the heart of the argument: the smallest saturated
> > acyclic alkane that cannot be made is a kind of alkane.
> Alkanes burn, bulk portions of them act as anaesthetics when inhaled, they react with chlorine radicals to yield chloroalkanes.
> If you admit non-existent alkanes as kinds of alkanes then you get nonsensical entailments such as non-existent alkanes being able to burn, to yield chloroalkanes when reacting with chlorine radicals, to anaesthetize people when inhaled.
Again, ontologies are not about representing particular instances, but
rather types of instances. If there would be an instance of a
non-synthesizable alkane, it would have all the properties you
describe. Therefore, the /type/ is a subclass. Only once you know what
the properties of the instances would be can you send a scientist to
look for them, try to synthesize them or investigate its properties,
possibility or impossibility (with respect to one particular
scientific theory).
Rob.
Chris Taylor
> ------------------------------------------------------------------------
>
>
> No virus found in this incoming message.
> Checked by AVG - www.avg.com
> Version: 9.0.839 / Virus Database: 271.1.1/3009 - Release Date: 07/16/10 07:35:00
Chris Taylor
Cheers, Chris.
________________________
chris....@ebi.ac.uk
http://mibbi.org/
David Osumi-Sutherland
"If the policy is not to accept an ontology until it is definitively proven..."
------------------------------------------------------------------------------
This SF.net email is sponsored by Sprint
What will you do first with EVO, the first 4G phone?
Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first_______________________________________________
Obo-discuss mailing list
Obo-d...@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/obo-discuss
Robert Hoehndorf
Hi Nicolas,
> > Yet this does not coincide with our intentions, because "absent arm"
> > is /used/ by scientists to refer to a state where there is no arm (and
> > no entity to refer to).
> That is not a good reason to adopt a term, for two reasons. The first is
> that there is a large body of "knowledge" used by people, including by
> "scientist" that is actually wrong. If you look at 10 textbook on
> neurobiology, 9 will mentioned that the action potential is generated at
> the level of the axon hillock (and this is what I leant in high school, and
> again in university). Nevertheless this is wrong. Any specialist will tell
> you that the action potential is initiated in the axon initial segment, and
> not the axon hillock. But the important information here is that each time
> I explained to someone what was the axon hillock, the initial segment and
> why was the axon potential initiated at the level of the second, this
> explanation was well received and the person changed her internal
> representation. My point here is that scientists are eager to learn and
> improve. And that brings me to the second point which is that one of the
> missions of ontologies in biology is to help people thinking clearly. And
> that works. Over the last decade, OBO helped me A LOT in properly
> formulating my understanding of the world, and this influenced my research.
> In summary I really disagree with the need of incorporating classes just
> because "people" use them. But designing clean ontologies can help people a
> lot, and they will gladly accept that.
Okay, so maybe you want to build ontologies to educate people and
change their thinking so that is fits one particular view on the
world. But OBO Foundry ontologies like the GO were developed to solve
a different problem (if I understand GO correctly): to specify the
meaning of terms that were already in use by scientists, that were
already used to annotate data in databases, to standardize the terms
that scientists used and needed to describe their research results (in
the model organism databases). And a "sorry, your term does not fit in
my ontology/philosophy" is not going to solve the problem.
If clinicians use "absent arm", have databases and health records that
include this term, and they want to build an ontology so that their
database can interoperate with those of other clinics, it seems
awkward to tell them that there is a kind of phenomenon that they must
first stop talking/thinking/researching about -- although it clearly
had its use in their database. Instead, the job of an ontologist could
be to analyze how "absent arm" is used in the particular situation,
and provide a consistent definition that reflects the use of the term
and is consistent within the ontology (and then it would not be a kind
of arm, but "not having an arm as part" or something similar).
This is not to say that every term ever used by people has to be
included in some ontology. But if, in the domain of your ontology, a
kind of phenomenon is refered to by the ontology's users, and the
users want to communicate about this kind of thing or represent
information about it in databases, then your ontology should include
this kind (at least if your ontology is supposed to standardize the
terms and their meaning in a domain so as to facilitate
interoperability and communication).
> > scientists /means/ when she uses "absent arm". I think it would be
> > along the lines of "not having an arm as part" or "Quality of things
> > that have no arm as part", but this is a debate for another time.
> Yes (for the last part) and no (for the first). As a parenthesis, let-me
> say that we should distinguish between the ontology and the way it is
> exposed to the end-user. Reversing your reasoning, we could generate an
> "absent arm" at the level of the user display based on the "not having an
> arm as part".
To me, "absent arm" is just a string. It is the job of the ontological
analysis to make its meaning explicit.
> The problem here is the Russel element/set problem. You have elements and
> sets. And they belong to different realms. An absent part does not exist.
But this does not mean that there is not a class of things which can
have the /label/ "absent arm", and this class of things is defined so
that it reflects the way that "absent arm" is used within a domain. It
may refer to situations of absence, to entities without an arm, or
similar. In the element/set example, ontology is about the sets, not
the elements. A class/type in an ontology is, in a way, an intensional
set (a set with a meaning attached to it). It does not even have to
have elements. And this still does not mean that "absent arm" is
either a set of arms (which are absent) or the empty set. It can still
be defined as the set of things which have no arm as part, or anything
else. There is no a priori method to formalize "absent arm".
Rob.
Adam M. Goldstein
Hello,Yet this seems at the heart of the argument: the smallest saturatedacyclic alkane that cannot be made is a kind of alkane. Otherwise thestatement (and a scientific paper about it) would not make sense(because it would be the "smallest molecule that cannot be made thatcannot exist". It is all about what problem ontologies are supposed tosolve: will they be catalogues of things which are believed to existby a certain group of scientists (all following a single scientifictheory), or do ontologies solve a communication and integrationproblem.
But, trivially, a "molecule which cannot be made" is not a molecule! It
does not have a weight or any other properties, it cannot be placed in a
bottle or on my hand, and it will never be a part of a drug or have any
kind of bioactivity... etc.
For_all_x : not-Fx which is the same same as (x)(~FX) for those familiar with that notation.
not-There_exists Fx, i.e., ~(Ex)Fx, where "E" is the existential quantifier.
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
Associate Editor
Reviews Editor
Evolution: Education & Outreach
http://www.springer.com/life+sci/journal/12052
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
Robert Stevens
The BFO version of realism is one such approach and I'd hesitate to say that the rigour thus provided is wholly bad -- it certainly isn't. it does seem to me, however, that the BFO approach at times does obfuscate rather than enlighten.
Modelling reality is obviously very appealing to scientists. we do, however, it seems have to model *only* instances in reality via *only* universals -- categories for those instances that themselves that have an existence independant of human thought. I do sometimes have problems in knowing when a universal does exist (and I know I'm not alone in this). the animals Michel, David, chris, Rob, janna and so on are the n real instances of the n+1 entity "Person" that also presumably exists. Dogs and cats have instances and presumably also have universals. The set (dog or cat) has instances, but no universal -- yet it seems to me that I might well wish to put this kind of category in my ontology to get the descriptions I wish.
yet it seems that I am often in need of, or willing to accept approximations that allow me to describe what we know and the science by which we know it.
BFO itself takes (or did when I last looked) a Newtonian view of the world. The physicists tell me this is not reality, but it is good enough to describe most of that we wish to describe. So, BFO is not realist as far as I can tell, but it doesn't bother me.
One of the oft visited topics in BFO is that we cannot have qualities that inhere in occurants or processes (again, last time I looked). thus velocities of reactions are rather tricky. Reactions may not in reality have velocities (I'm not in a position to say), but the proscription of qualities inhering in occurants does stop me saying what I think is necessary to describe stuff. I note that Pato does have qualities of occurants....
Numbers, especially zero, seem like a problem. units, another human fiction seem rather necessary to describe that which we wih to describe. I understand that the approach taken is to have all of this kind of stuff as information artefacts (well numerals and labels for units etc), but this seems rather obfuscatory to me. I'd be willing to ignore the reality in favour of some simpler fiction.
We also have many scientific conveniences -- my friend Phil Lord offers the nice example of colour models as a means by which we describe colour. Describing the reality of colour -- transmission, reflection, the perceiver (I might see a flower as Red, but a bee sees something different; primates have three colour vision, but other mammals two colour vision) seems like rather hard work. so I, like Pato, would use the non-real quality as described by a colour model. the model exists, but is presumably in reality some kind of information artefact -- again, I'm willing to accept the "fiction" of the colour model as a quality, not an information artefact.
I was also recently thinking about pH -- a nice human conception of a negative logarithm that describes proton concentration -- again, presumably an information artefact and not a quality). To capture the pH of something would mean I have more hoops to jump through than I really think are necessary -- unless we decide that making scientists undertstand the reality of units is what we need.
this got a bit long, but I hope goes some way to outline my "entry requirements" for a class to get into an ontology.
Robert.
Adam M. Goldstein
Hello all,
As regards the following quote Rob cites, I would like to point out that a
planned or predicted molecule is definitely not a drug product, and would
rightly be excluded from any ontology of drug products. Drug products are
those things that have brand names, packaging, dosage recommendations,
package inserts listing side effects and have been approved for various
indications by a legal body which has the power to endorse drugs.
I cannot imagine any pharmaceutical application use case for which an
ontology of *drug products* would benefit from the inclusion of millions
of not-yet-synthesised specifications from a combinatorial chemical
library.
Escitalopram is an enantiopure compound of the racemic mixture citalopram, used for the same indication, and for that reason it required less investment and less time to develop. Two years after escitalopram's launch, when the patent on citalopram expired, the escitalopram sales successfully made up for the loss. On May 23, 2006, the FDA approved a generic version of escitalopram by Teva. On July 14 of that year, however, the U.S. District Court of Delaware decided in favor of Lundbeck [the owner of the citalopram patent] regarding the patent infringement dispute and ruled the patent on escitalopram valid.
Nicolas Le novère
> The attached illustration is from _Quirks of Human Anatomy: An Evo-Devo
> Look at the Human Body_, by Lewis I. Held, Jr., p. 3.
>
> I think we all agree that the white circle belongs in OBO-compliant
> ontologies, as stated by the criterion, and the black circle is flat
> out, as they violate the laws of physics.
Actually this is even more complex than that. The extinct circle may
contain classes for which there are no instances at the moment. An example
would be the named common ancestors of species for which we have current
artifacts (e.g. living instances or fossils). Would-that be accepted in
OBO? If not, there is no hope of creating an OBO taxonomy based on the
phylocode. If yes, I see a temporal dis-symmetry, where classes for which
there were instances but not anymore are accepted, while classes for which
there are not yet instances (the pharmaceutical drug) is not accepted.
> What I am asking is what is the policy on things that fall in the grey
> circle in our current state of knowledge. They are physically possible,
> and at the time those ideas flourished, they were the best explanation
> available, but they have been supplanted by later knowledge. Sailors'
> reports of mermaids turned out to be manatees and dugongs, dragons are
> reptiles, and so forth. The original conceptualisations turned out to be
> false in the end, but at the time, they represented the best available
> knowledge.
Indeed
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
Nicolas Le novère
> My impression has always been that the Foundry are not interested in
> ontologies in which classes are 'theoretical' and have no known
> instances in reality (at present).
Hmmm. Who is the Foundry? It seems that the current discussion shows that
among OBO Foundry contributors, there is a perception that the limits of
the Foundry are maybe not defined precisely enough, in the current case
what is reality, or that those limits are maybe not that consensual.
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Colin Batchelor
> I don't know, the company wants to corner the market and tries to
> claim a patent for all drugs of a certain class, or derived from
> a particular precursor compound?
There are claims like that but they tend to be specified mereologically (through Markush structures) rather than through listing all of them.
Certainly you have to prove that at least all the bits (that one might assemble in combinatorially many ways) in your claim exist, so all is well.
Colin.
--
Dr Colin Batchelor MChem CChem CSci,
Senior Informatics Analyst,
Royal Society of Chemistry, Thomas Graham House, Cambridge, UK CB4 0WF.
t: +44 1223 432280 e: batch...@rsc.org
DISCLAIMER:
This communication (including any attachments) is intended for the use of the addressee only and may contain confidential, privileged or copyright material. It may not be relied upon or disclosed to any other person without the consent of the RSC. If you have received it in error, please contact us immediately. Any advice given by the RSC has been carefully formulated but is necessarily based on the information available, and the RSC cannot be held responsible for accuracy or completeness. In this respect, the RSC owes no duty of care and shall not be liable for any resulting damage or loss. The RSC acknowledges that a disclaimer cannot restrict liability at law for personal injury or death arising through a finding of negligence. The RSC does not warrant that its emails or attachments are Virus-free: Please rely on your own screening.
------------------------------------------------------------------------------
Ravensara Travillian
The elephant in the kitchen here is epistemology.
> "If the policy is not to accept an ontology until it is definitively
> proven..."
I don't think that the job of science is to prove things. It is to
provide good evidence for them. Proof is an impossibly high bar to
set.
But, at any one time there are very many things for which there
is a strong scientific consensus based on extremely good evidence.
Now, this means there is a grey area - what things are too contentious
to go into a realist ontology right now? What counts as good evidence
for a particular assertion? All I can say is, use your judgement as a
scientist. If you're unsure, ask outside experts.
[Raven: It strikes me that you should not use the LCA approach for
something as ancient and contentious as the relationship of
echinoderms to vertebrates. But it could surely be safely used to
define the homology grouping 'mammalian liver'.]
Perhaps a more neutral way to have this argument is to talk about
ontology use cases. I know that, for the anatomy ontology I build,
mixing in classes that lack instances
and assertions we know to be
incorrect
would be disastrous. A user querying are ontology right now
can find a list, for example, of all the classes of neurons of some
specified lineage that, in wild-type flies, innervate the antennal
lobe. Now, the lineage tracing studies and evidence for innervation
patterns that this is based on are not perfect. There have been
mistakes in these in the past, and there are probably a few still.
But the results of the query are useful and link back to the
literature so that scientists can check the evidence and use their own
judgment. The results themselves can then be used to retrieve
annotations. So, for example, we can ask what genes are expressed in
neurons of the specified lineage that innervate the antennal lobe.
I'm not hostile to the use cases in hypothesis testing that Robert and
Michel are fond of, but we need to be very careful that, in allowing
these, we don't screw up use cases like the ones I've described, that
many of us depend on.
Adam M. Goldstein
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first_______________________________________________
> Obo-discuss mailing list
> Obo-d...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
Associate Editor
Reviews Editor
Evolution: Education & Outreach
http://www.springer.com/life+sci/journal/12052
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
------------------------------------------------------------------------------
Chris Mungall
On Jul 16, 2010, at 5:23 AM, Robert Hoehndorf wrote:
> On Fri, 2010-07-16 at 11:58 +0100, James Malone wrote:
>
> Hi,
>
>> My impression has always been that the Foundry are not interested in
>> ontologies in which classes are 'theoretical' and have no known
>> instances in reality (at present). Therefore, the work Michel and Rob
>> describe is likely out of scope. However, just because it is out of
>
> I would argue that ontologies like OBI or IAO, which attempt to
> provide support for what scientists do (collect data, create theories,
> make predictions from these theories, verify them by experiment),
> would then be out of the OBO Foundry's scope, too.
>
> I am not sure, but I think it is quite possible that even the GO has
> classes without instances in reality, or at least without any /known/
> instances. Good candidates are leaf classes without any
> annotations. Classes which are included because they were predicted,
> conjectured, hypothesized, and scientists are starting just now to
> conduct experiments to learn more about them (if they have
> instances). Maybe somebody with a better knowledge of biology or the
> GO than me can attempt to find one.
Come one Rob. There's pretty good reason to believe that every class
in GO has instances. See David's email - it doesn't matter if an
instance has been directly involved, use your scientific judgment. If
anyone finds something like binding to an impossible alkane we'll
obsolete it immediately.
David Osumi-Sutherland
On 16 Jul 2010, at 16:06, Ravensara Travillian wrote:
> insisting on instances sounds like--to me--to insist on proving
> every single entity existed
I think its just insisting that we have good scientific reasons to
believe that instances of the class exist/can exist/have existed. If
someone at the LHC claims to have found an instance of Higg's Boson,
what they'll really be saying is that the all their measurements of
bubble trails (or whatever methods they now use for these experiments)
are consistent with the presence of a particle with the properties
predicted for Higg's Boson. I don't see this as proof - someone else
might come along with a different and highly plausible interpretation
of the same data. But it provides a good reason to believe that
instances of Higg's boson exist. I don't see assertions about
evolutionary history as so massively different from this. We can't
go back in time and identify instances of common ancestors. All we'll
ever have is the clues of extant species and fossils. Still, there
are many assertions we can make about evolutionary history, based on
this evidence, that are boringly uncontentious.
There seems to be general agreement here that the OBO Foundry rules
need to be clarified - particularly with respect to time. But I don't
see any prospect of coming up with guidelines for what types of
evidence counts. Nor should there be. That's the job of science.
Finally - can we take this discussion of LCAs off the main list. I
think the details are a bit of a branch from the main thread here.
I'll try to reply to some of your points directly.
Cheers,
David
David Osumi-Sutherland, PhD
Ontologist / Curator
Virtual Fly Brain / FlyBase
Department of Genetics
University of Cambridge
Downing Street
Cambridge, CB2 3EH
UK
+44 (0)1223 333 963
Peter Midford
Please add me to the spin-off discussion.
Peter
Peter E. Midford
Phenoscape Ontology Curator
Peter....@gmail.com
Peter Karp
(a) The paper by Dumontier and Hoehndorf makes a number of excellent
points. It will of course be interesting to see what Barry et al
have to say in response.
(b) The OBO Foundry Principles are so unclearly expressed that they
are not deserving of all the key pokes they have generated. The
instantiability principle is one example of such a vaguely formulated
principle (in fact, it is rather ironic that a group concerned with
ontologies has formulated such unclearly stated principles for ontologies).
Most of the principles on that page require further discussion,
explanation, and justification, and I recommend that the OBO Foundry
group create an accompanying web page that provides a more in-depth
discussion of each principle. Examples would be particularly helpful --
"near misses" are always a helpful way of clarifying definitions.
For example, it would be helpful to know what Barry or others think
are possible ontology terms that people might want to define that
should not be defined because they do not correspond to instances
in reality. What does it even mean to correspond to an instance in
reality? Is a regulation event an instance in reality? Is an
experiment plan an instance in reality? Is the process of transcription
an instance in reality? Or are these entities simply cognitive
constructs? (No doubt these issues are discussed in Barry's
papers, but the OBO Foundry principles should be comprehensible
on their own.)
P
Suzanna Lewis
In answer to Nic's question. The OBO-Foundry includes all the people interested in this discussion who and contribute to this forum (including critiques).
More concretely, the OBO-Foundry is comprised of the editors who maintain the ontologies by defining classes and the relationships between them and maintain them when scientific advance indicates that changes are needed (e.g. example in fly anatomy from David O-S). This is primarily with the objective of supporting scientific research, including generation of new hypotheses. The editors work cooperatively to make the suite of ontologies non-redundant so that they complement one another and can be used in combination. This group is really the heart of the OBO-Foundry, and are motivated by an interest in biology/biomedical questions, and philosophical only to the extent that it is a tool for logical reasoning that can inform their biological research.
Operationally (all with donated time), there are a few of us who set up mailing lists, help maintain repositories, organize meetings and workshops, assist in coordination between the different ontology development groups, etc.. This includes myself, Michael Ashburner, Barry Smith, Alan Ruttenberg, Richard Scheuermann, Chris Mungall, and Susanna-Assunta Sansone.
Historically the OBO-Foundry got started because people who needed an ontology for their research purposes would e-mail us individually with requests for help, largely because of the GO's success. It seemed clear there was a need for a forum and for the development of standards. As a group we are working to facility this discourse and assist in improving the practical usage of for biological&biomedical research. As Janna said "ontologies are works in progress, continuously evolving as the relevant domain science progresses."
----
Last, members of OBO-Foundry (i.e. content editors) include evolutionary biologists and we do include classes for which there is scientific evidence that instances did exist at some time in the past. For example, the use of GO classes to annotate the inferred function of ancestral nodes in protein family trees.
Cheers, Suzanna
Nicolas Le novère
> On 16 Jul 2010, at 16:06, Ravensara Travillian wrote:
>
>> insisting on instances sounds like--to me--to insist on proving
>> every single entity existed
>
> I think its just insisting that we have good scientific reasons to
> believe that instances of the class exist/can exist/have existed.
"can exist", youhou! Let's replace by "we have good scientific reasons to
believe that instances of the class exist/can exist/have existed/will
exist", and we have an emerging consensus for the "extended realist position".
(Rob, I believe several discussions are covered by this thread, let's
serialise)
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Chris Mungall
On Jul 16, 2010, at 10:42 AM, Nicolas Le novère wrote:
> On 16/07/10 17:35, David Osumi-Sutherland wrote:
>
>> On 16 Jul 2010, at 16:06, Ravensara Travillian wrote:
>>
>>> insisting on instances sounds like--to me--to insist on proving
>>> every single entity existed
>>
>> I think its just insisting that we have good scientific reasons to
>> believe that instances of the class exist/can exist/have existed.
>
> "can exist", youhou! Let's replace by "we have good scientific
> reasons to
> believe that instances of the class exist/can exist/have existed/will
> exist", and we have an emerging consensus for the "extended realist
> position".
This seems fine to me.
Suzanna Lewis
Chris Mungall
On Jul 16, 2010, at 3:58 AM, James Malone wrote:
> I confess I am not expert on unicorn anatomy though somebody told me
> the
> "surprises" component in the diagram is more a posterior element...
>
> I find this discussion is useful and I'm grateful for the original
> posting. The OBO Foundry, as Nicolas has rightly pointed out, is
> seen by
> many as a standards body and as such should also be open to scrutiny -
> just as the ontologies that sit under its wing should. I know Barry et
> al. have published a large body of work describing problems with
> ontologies before Realist approaches and to the extent that I think
> things have improved dramatically over the last few years, I agree
> with
> him. Whether or not that is because of BFO I honestly do not know
> and it
> is a reasonable, scientific question to ask whether we gain or lose by
> this requirement; we should attempt to answer this objectively I
> believe. I am grateful to Barry et al for their contribution to the
> field and I am grateful to Michel, Rob et al for adding some
> discourse;
> we are better for it.
>
> My impression has always been that the Foundry are not interested in
> ontologies in which classes are 'theoretical' and have no known
> instances in reality (at present). Therefore, the work Michel and Rob
> describe is likely out of scope. However, just because it is out of
> scope for OBO, if indeed that is the case, does not make this work any
> less valuable to the community. It is possible that in the future some
> of the most useful classes will have first started life in an ontology
> such as the one Michel describes and as such, should not be
> discouraged.
Well put James.
I agree with all this except my own personal opinion my be even more
liberal with respect to physically possible compounds that just
haven't been synthesized yet. I certainly don't think CHEBI should
have the responsibility of taking these on board, but I'm not sure
they should be excluded a-priori on philosophical grounds. I would
punt this on to the "does the ontology have a plurality of independent
users" principle.
I remain skeptical about impossible-to-instantiate compounds but I'm
willing to be convinced. If philosophical dogmatism is obstructing
scientific work or data integration then it's clear we should abandon
the dogmatism.
> Cheers,
>
> James
>>> ------------------------------------------------------------------------------
>>> This SF.net email is sponsored by Sprint
>>> What will you do first with EVO, the first 4G phone?
>>> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
>>> _______________________________________________
>>> Obo-discuss mailing list
>>> Obo-d...@lists.sourceforge.net
>>> https://lists.sourceforge.net/lists/listinfo/obo-discuss
>>>
>>
>> David Osumi-Sutherland, PhD
>> Curator/ Ontologist
>> FlyBase / Virtual Fly Brain
>> Department of Genetics,
>> University of Cambridge,
>> Downing Street,
>> Cambridge, CB2 3EH, UK
>> Tel: +44 (0)1223 333 963
>> Fax: +44 (0)1223 766 732
>>
>>
>> ------------------------------------------------------------------------------
>> This SF.net email is sponsored by Sprint
>> What will you do first with EVO, the first 4G phone?
>> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
>> _______________________________________________
>> Obo-discuss mailing list
>> Obo-d...@lists.sourceforge.net
>> https://lists.sourceforge.net/lists/listinfo/obo-discuss
>>
>
>
> --
> European Bioinformatics Institute,
> Wellcome Trust Genome Campus,
> Cambridge, CB10 1SD,
> United Kingdom
> Tel: + 44 (0) 1223 494 676
> Fax: + 44 (0) 1223 494 468
>
>
Gary Merrill
Let me beg your indulgence and begin with a story. Many years ago when I was
a professor of philosophy (with some effort you can figure out where), one
year we were faced with hiring some new faculty. The department was
somewhat factional and there was an overt concern about hiring someone who
was “too analytic” since that would push what was viewed as the factional
balance down a road many did not want it to go. Accordingly, a criterion of
acceptability for any new hire was proposed and voted on by the graduate
committee as a whole. The criterion was this: that anyone offered the
position must be “open to the transcendent”. Although it was deemed
impossible to define this phrase, or even to characterize it in an
unambiguous manner, the proposal passed (by unanimous vote!) and the
criterion was adopted. I do not believe that it was ever invoked to exclude
a candidate. How could it be? No one knew what it meant. But it made a
significant portion of the faculty feel a lot better. I voted for it myself
because I couldn't see any harm in it, and I cherished the idea of being,
myself, certified as open to the transcendent. You just never know when
something like that can come in handy. But that was a philosophy faculty
committee, and not science.
I take this truth to be self-evident: that in order for a criterion to be
of value and use, there must be an objective way of determining when and how
it applies. Otherwise, you don't have a criterion. You've got an
expression of emotion, or a statement of some kind of ideology, or some
such.
For reasons similar to my openness to the transcendent, I am inclined in
part not to be overly concerned with the appeal to “realism” in the OBO
Foundry Principles. (Though, having grown older if not wiser, I would not
have voted for such a criterion.) So far as I can see, this appears only in
two clauses in the “Additional Principles” (clauses 6, and 10). There is
not, so far as I can see, any way of determining under what conditions a
term does “correspond to something in reality” (clause 6) or when something
is one of the “types in reality”. Indeed, one might suspect that the
inclusion of these clauses may be the result of some sort of committee
compromise along the lines of openness to the transcendent.
Consequently, my own problem with these particular Foundry Principles is not
that I regard them as wrong (or false, or incorrect), but rather that I
regard them as quite literally senseless – in the sense that the operative
phrases in them genuinely have no identifiable sense. They are meaningless
… devoid of sense and meaning. ( Cf. Python, Monty "The Dead Parrot", 1969.
I submit that similarly these are dead principles.) And thus I can't
imagine how they could ever be applied. And if people like having literally
senseless principles among their criteria, then there is probably no harm in
this so long as no attempt is ever made to invoke them.
Now comes the rub for those (a part of my transcendently open self included)
who would see such things as harmless. They are thus also useless and
senseless, and it's probably not a good idea to adopt useless and senseless
principles in what appears to be a scientific endeavor. But perhaps that's
a matter of taste. The problem comes if anyone ever actually does try to
invoke them and actually makes a decision or casts a vote based on a
proposed OBO ontology failing such criteria. This would, I suspect, result
in a rather spirited dispute that could not be resolved in anything
approaching an objective manner. ("My term does refer to something in
reality" ... "Does not!" ... "Does to!" ... etc.)
Under what conditions might this happen? Well, the history of the
literature suggests that it might happen if someone proposes a “concept
based” ontology rather than an ontology that acknowledged “universals” as
the designata of its general terms. That is only speculation, of course,
but it leads to the question of exactly what a case would look like that
failed to conform to these “realist” principles. Would an ontology based
on, for example, Cyc be rejected because it was concept based and
consequently its terms fail in “corresponding to something in reality” or
referring to “types in reality”? Would the same be said of an ontology
based on something like Cocchierella's “conceptual realism” or the work of
Zalta or Guarino? It is difficult to believe that such would be an intended
consequence of these criteria. But then what? A measure of the
senselessness of these phrases in the OBO Principles is that if they were to
be removed, it is pretty much impossible to see what effect that would have
on anything. It is hard to fault these proposed principles because,
currently, it is impossible to understand them.
Now I could be wrong about this. It could be that the OBO does (or does
intend to) impose a meaning on these principles that lends them sense and
anchors the apparent references to reality in … uh … reality. (I confess
that it is not at all clear to me exactly how you are supposed to do that.)
And in that case, we would presumably be able to identify genuine
consequences of applying the principles, and determine whether and to what
degree these are beneficial or harmful. For now, I don't see that we can do
so. There is, from my perspective, no way to resolve a dispute here because
there doesn't appear to be a substantive dispute – owing to the lack of
sense to the principles in question. In that case, the only dispute
concerns whether to adopt senseless principles. Being open to the
transcendent, I am happy to leave that decision to others; but I would not
be inclined to adopt such principles myself -- at least in something I'm
pretending would be real science.
(Note that the points I make here have absolutely nothing to do with whether
realism, in whatever form you favor it – belief in material objects, in
Platonic forms, in Aristotelian universals, in tropes, in ghosts and
spirits, in cold fusion, etc. – is a good position to adopt. So these
points are largely, if not wholly, independent of anything that is argued in
either of the papers cited by Michel and Rob.)
--
View this message in context: http://obo-discuss.2851485.n2.nabble.com/Ontological-Realism-and-OBO-Foundry-Criteria-tp5293729p5303384.html
Sent from the OBO Discuss mailing list archive at Nabble.com.
Adam M. Goldstein
> seconded
>
> On Jul 16, 2010, at 1:53 PM, Chris Mungall wrote:
>
>>
>> On Jul 16, 2010, at 10:42 AM, Nicolas Le novère wrote:
>>
>>> On 16/07/10 17:35, David Osumi-Sutherland wrote:
>>>
>>>> On 16 Jul 2010, at 16:06, Ravensara Travillian wrote:
>>>>
>>>>> insisting on instances sounds like--to me--to insist on proving
>>>>> every single entity existed
>>>>
>>>> I think its just insisting that we have good scientific reasons to
>>>> believe that instances of the class exist/can exist/have existed.
>>>
>>> "can exist", youhou! Let's replace by "we have good scientific
>>> reasons to
>>> believe that instances of the class exist/can exist/have existed/will
>>> exist", and we have an emerging consensus for the "extended realist
>>> position".
>>
>> This seems fine to me.
>>>
Suzanna seconds this above, and I don't want to be contrary, but I think that this is going to be too hard to really get right, and that the realists are giving in way too easily here. "Can exist" is a modal notion; and surely things that will exist, but don't now aren't real, right? Moreover this is all couched in a "reason to believe." Who has reason to believe? Me? Somebody really smart? The OBO Foundry administrators? Not to be flip about it but this is going to be really hard to make sense of.
It seems to me that the realist position articulated (some people will say very poorly) in the context of the OBO is meant to reflect the idea that ontologies model *the domain under study* and not, for instance, what people think or believe about it, how they conceptualize it, and so on. The ontology might express *what* people think, but this is just an attitude someone might have about the ontology, not part of its content. If physics is about atoms, then the ontology should model the world of atoms, and it should tell us about what happens in that world if certain states of atoms represented in the model obtain. Similarly for ontologies and machine reasoning using ontologies.
If the proposal above is accepted, then OBO ontologies will be open to including possibilities of atoms, future atoms, atoms that have now been destroyed in tests of hydrogen bombs, and the like. These things are definitely not kinds of atoms. If someone adopts this view, he or she is a realist extraordinaire! It doesn't go quite so far as David Lewis' "Any world that can exist, does." I don't think this is what the OBO realism is intended to capture.
This has been an engaging discussion and I feel that it was well worth the time I spent following it. My laptop battery is almost dead so maybe it's time to call it a day...
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
Associate Editor
Reviews Editor
Evolution: Education & Outreach
http://www.springer.com/life+sci/journal/12052
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
------------------------------------------------------------------------------
Barry Smith
which can be recommended to scientific researchers as capturing the
terminological content of established science in a given area. Such
reference ontologies, we assume, will allow multiple different kinds
of application ontologies to be more successfully created, including
application ontologies involving provisional assumptions (which we
assume will be marked as such by means of appropriate evidence codes).
All ontologies are subject to constant revision, and this includes
the OBO Foundry itself.
BS
At 05:23 AM 7/16/2010, Ravensara Travillian wrote:
>Since I did not get an answer to the question I posed, perhaps I
>need to restate it.
>
>The attached illustration is from _Quirks of Human Anatomy: An
>Evo-Devo Look at the Human Body_, by Lewis I. Held, Jr., p. 3.
>
>I think we all agree that the white circle belongs in OBO-compliant
>ontologies, as stated by the criterion, and the black circle is flat
>out, as they violate the laws of physics.
>
>What I am asking is what is the policy on things that fall in the
>grey circle in our current state of knowledge. They are physically
>possible, and at the time those ideas flourished, they were the best
>explanation available, but they have been supplanted by later
>knowledge. Sailors' reports of mermaids turned out to be manatees
>and dugongs, dragons are reptiles, and so forth. The original
>conceptualisations turned out to be false in the end, but at the
>time, they represented the best available knowledge.
>
>So, I will state my questions more clearly, in the hope of obtaining answers:
>
>Is the realism criterion to be read literally?
>
>If so, does that mean a binary assumption of how morphospace is to
>be carved up? Does it assume only the white circle and the black
>circle, and nothing else?
>
>If that is the case, does that mean that any evo-devo ontology I
>develop with provisional assumptions about LCAs, to be revised in
>light of later evidence [thus, falling in the grey circle], is by
>definition, not eligible for inclusion in OBO?
>
>If it is not to be read literally, then can I take Chris'
>description of accepting an ontology and then revising it in light
>of later knowledge as an accurate statement of official OBO policy?
>
>If that is the case, then what is the policy regarding earlier
>accepted ontologies that--only in light of later knowledge--are
>shown not to be OBO-compliant, yet were accepted?
>
>(And Held points out that the narwhal and the Indian rhino are
>technically unicorns, as well, since we keep coming back to that
>imagery. Just sayin'.)
>
>Cheers,
>
>Raven
>
>--
>Ravensara S. Travillian
>Vertebrate Anatomist/Bioinformatician
>European Bioinformatics Institute,
>Wellcome Trust Genome Campus,
>Cambridgeshire, CB10 1SD,
>United Kingdom
>Tel: + 44 (0) 1223 494 553
>Fax: + 44 (0) 1223 494 468
>Content-Type: image/png; name="morphospace.png"
>Content-Disposition: attachment; filename="morphospace.png"
>X-Attachment-Id: f_gbosw5j10
Barry Smith
>My take on this discussion is
>
>(a) The paper by Dumontier and Hoehndorf makes a number of excellent
>points. It will of course be interesting to see what Barry et al
>have to say in response.
>
>(b) The OBO Foundry Principles are so unclearly expressed that they
>are not deserving of all the key pokes they have generated. The
>instantiability principle is one example of such a vaguely formulated
>principle (in fact, it is rather ironic that a group concerned with
>ontologies has formulated such unclearly stated principles for ontologies).
>
>Most of the principles on that page require further discussion,
>explanation, and justification, and I recommend that the OBO Foundry
>group create an accompanying web page that provides a more in-depth
>discussion of each principle.
I believe that we have been trying to do this, in very many papers,
lectures. videos, etc. But we can of course try harder.
> Examples would be particularly helpful --
>"near misses" are always a helpful way of clarifying definitions.
>For example, it would be helpful to know what Barry or others think
>are possible ontology terms that people might want to define that
>should not be defined because they do not correspond to instances
>in reality.
unicorn
mermaid
leprechaun
absent nipple
absent leg
cancelled performance
entity that is simultaneously an instance of Heart failure, Tooth
decay, and Pregnancy
single-celled mammalian organism
> What does it even mean to correspond to an instance in
>reality?
An instance is something that exists in space and time. Often
instances can be observed or measured. For a referring expression to
correspond to an instance means nothing more sophisticated than for
it to refer to or name that instance. E.g. 'Peter Karp' corresponds
to Peter Karp. Part of the problem is that we are dealing with issues
so basic that there is very little further that we can say that would
illuminatingly explain them.
> Is a regulation event an instance in reality?
Yes
> Is an
>experiment plan an instance in reality?
Yes
> Is the process of transcription
>an instance in reality?
Any given transcription process occurring in some given place and
time is an instance in reality; an instance of the type transcription process?
> Or are these entities simply cognitive
>constructs?
If they exist independently of scientists' descriptions of them then
they are not cognitive constructs.
>(No doubt these issues are discussed in Barry's
>papers, but the OBO Foundry principles should be comprehensible
>on their own.)
We are, indeed, doing our best.
BS
Barry Smith
Ceusters W, Elkin P, Smith B.
<http://ontology.buffalo.edu/medo/NegativeFindings.pdf>Negative
Findings in Electronic Health Records and Biomedical Ontologies: A
Realist Approach. International Journal of Medical Informatics 2007;
76: 326-333.
Adam M. Goldstein
On Jul 16, 2010, at 17:35, Barry Smith <phis...@buffalo.edu> wrote:
>> Most of the principles on that page require further discussion,
>> explanation, and justification, and I recommend that the OBO Foundry
>> group create an accompanying web page that provides a more in-depth
>> discussion of each principle.
>
> I believe that we have been trying to do this, in very many papers,
> lectures. videos, etc. But we can of course try harder.
>
I find the documentation in the ontologies in the comments and definition field very useful.
I think something like it would be useful for people: the formal definition, a brief explanation, at least one example of each showing a case that meets the criteria, and one that doesn't, with a brief explanation why.
The papers and other materials are excellent resources---but it can hard to find the information one needs. For instance in the paper about ontologies for dynamic geography, linked I think from the BFO home page, there is some really good basic stuff---but who would think to look there for an overview.
I know that people are not jumping up and down to write documentation, but for my part I will try to do do something like this in my own ontologies...
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
Chris Mungall
On Jul 16, 2010, at 2:35 PM, Barry Smith wrote:
>>
>> Examples would be particularly helpful --
>> "near misses" are always a helpful way of clarifying definitions.
>> For example, it would be helpful to know what Barry or others think
>> are possible ontology terms that people might want to define that
>> should not be defined because they do not correspond to instances
>> in reality.
>
> unicorn
> mermaid
> leprechaun
> absent nipple
> absent leg
> cancelled performance
> entity that is simultaneously an instance of Heart failure, Tooth
> decay, and Pregnancy
Some of these could be rewritten as perfectly acceptable classes - the
condition of having a missing nipple, or a torso that is missing a
nipple.
Some of the others could be excluded on other grounds without invoking
realism.
Nevertheless the realism principle is still a useful first line of
defense. Asking someone to describe an instance of a problematic class
is a useful tool. But I suspect it's unlikely we'll have to invoke
this clause because so few bio-ontologies that pass the other criteria
have any real reason to represent classes that violate the principle.
Rob and Michel's impossible-compound case presents an interesting
challenge - I believe them that their use case is important and I want
to find a way to bring in what they're doing in a way that's
satisfactory to everyone.
Another example for Peter, taken from the venerable concept-based Cyc
ontology (nothing to do with Peter's Eco/MetaCyc) is the class "Heart,
locus of feelings"
The metaphorical place where all of a sentient being's
occurrent emotional experiences are said to be felt. In
metaphorical terms, the body's container of emotions that we
commonly associate with the hear or bosom. This container
underspecified is considered a part of the mind, itself being
a container of all thoughts. Emotions are associated with the
heart or bosom because many strong emotions manifest
themselves, in part, as physical feelings in one's chest area.
And also:
http://sw.opencyc.org/2008/06/10/concept/Mx4rvhn_I5wpEbGdrcN5Y29ycA
No disrespect to Cyc, conceptual entities such as these shouldn't be
in the OBO Foundry.
> single-celled mammalian organism
mammalian zyogote?
Peter Karp
Things that are controversial. No one in biomedical informatics
would plan on including unicorns in their ontology. Where is the
principle actually helpful in resolving controversy?
What about encoding negation? If I do an experiment and find that
E. coli cannot utilize (say) maltose as a carbon source, can I
capture that in an OBO ontology? It's an important scientific
fact to capture, but I can't tell whether it would correspond to
reality in the OBO sense.
P
Wacek Kusnierczyk
> On Jul 16, 2010, at 2:35 PM, Barry Smith wrote:
>
>
>>> Examples would be particularly helpful --
>>> "near misses" are always a helpful way of clarifying definitions.
>>> For example, it would be helpful to know what Barry or others think
>>> are possible ontology terms that people might want to define that
>>> should not be defined because they do not correspond to instances
>>> in reality.
>>>
>> unicorn
>> mermaid
>> leprechaun
>> absent nipple
>> absent leg
>> cancelled performance
>> entity that is simultaneously an instance of Heart failure, Tooth
>> decay, and Pregnancy
>>
> Some of these could be rewritten as perfectly acceptable classes - the
> condition of having a missing nipple, or a torso that is missing a
> nipple.
>
>
Do you say that 'having a missing nipple' is perfectly acceptable while
'absent nipple' is unacceptable?
vQ
Nicolas Le novère
>> Some of these could be rewritten as perfectly acceptable classes - the
>> condition of having a missing nipple, or a torso that is missing a
>> nipple.
>>
>>
>
> Do you say that 'having a missing nipple' is perfectly acceptable while
> 'absent nipple' is unacceptable?
No, 'having a missing nipple' is not acceptable. I believe Chris slipped.
But "a torso that is missing a nipple" is fine.
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Wacek Kusnierczyk
> On 17/07/10 11:41, Wacek Kusnierczyk wrote:
>
>
>>> Some of these could be rewritten as perfectly acceptable classes - the
>>> condition of having a missing nipple, or a torso that is missing a
>>> nipple.
>>>
>>>
>>>
>> Do you say that 'having a missing nipple' is perfectly acceptable while
>> 'absent nipple' is unacceptable?
>>
> No, 'having a missing nipple' is not acceptable. I believe Chris slipped.
> But "a torso that is missing a nipple" is fine.
>
I guess 'lack of a nipple' is fine, too?
vQ
Janna Hastings
Let us imagine that we concede that empty types are useful to formulate in
purely logical terms, making no appeal to reality or the existence of
instances to guide our ontology development. Surely, even so, our
aspiration in chemical ontology is to encode, in the ontology, precisely
the physical rules of the feasibility of molecules as we discover them.
Encoding structural features as necessary and sufficient conditions on
classes is an active area of research in chemical ontology at the moment.
If we assume for the sake of argument that we could do this properly, and
encode the structural features of feasible molecules as necessary and
sufficient conditions on the class 'molecule', and also encode precisely
the structural features of this impossible alkane, then a reasoner would
be able to pick out the class as inconsistent if it is asserted as a
subclass of molecule. Surely we do not want to encourage a future
generation of ontologists to create inconsistent classes in their
ontologies?
Cheers, Janna
> On Fri, 2010-07-16 at 13:38 +0100, Colin Batchelor wrote:
>
> Hi,
>
>> > Yet this seems at the heart of the argument: the smallest saturated
>> > acyclic alkane that cannot be made is a kind of alkane.
>> Alkanes burn, bulk portions of them act as anaesthetics when inhaled,
>> they react with chlorine radicals to yield chloroalkanes.
>> If you admit non-existent alkanes as kinds of alkanes then you get
>> nonsensical entailments such as non-existent alkanes being able to burn,
>> to yield chloroalkanes when reacting with chlorine radicals, to
>> anaesthetize people when inhaled.
>
> Again, ontologies are not about representing particular instances, but
> rather types of instances. If there would be an instance of a
> non-synthesizable alkane, it would have all the properties you
> describe. Therefore, the /type/ is a subclass. Only once you know what
> the properties of the instances would be can you send a scientist to
> look for them, try to synthesize them or investigate its properties,
> possibility or impossibility (with respect to one particular
> scientific theory).
>
> Rob.
Nicolas Le novère
> Moreover this is all couched in a "reason to believe." Who has
> reason to believe? Me? Somebody really smart? The OBO Foundry
> administrators? Not to be flip about it but this is going to be really
> hard to make sense of.
But ... this is already the case. Every single term of GO is designed that
way. Actually, even for ChEBI, for which there is a possible onoe to one
experimental validation of the leaves, the nodes are just conceptual. And
this is where the realist position is untenable. Ontologies are inherently
about concepts and their relationships. Effectively, except for some
leaves, all the other terms are designed on a rationalist point of view
(and btw, as a rationalist, I am a bit fed-up of the unicorn story. For the
vast majority of the population a "rationalist" is someone who does not
believe/use the concepts of unicorns, father christmas, fairies and god.
The fact that many of us believe that what we are dealing with here are
productions of our minds in response of sensory input and past experience,
does not make us raving lunatics).
> It seems to me that the realist position articulated (some people will
> say very poorly) in the context of the OBO is meant to reflect the idea
> that ontologies model *the domain under study* and not, for instance,
> what people think or believe about it, how they conceptualize it, and so
> on.
? I am puzzled by that claim
GO:0023033 signaling pathway
Both "signaling" and "pathway" are conceptualisation. The same biochemical
reaction, let's say GTP->GDP+P can be "seen" as a metabolic reaction,
producing energy, or as part of heterometic G-protein signaling. "pathway"
is actually a concept that is deeply hurting any "pathway" database at the
moment, because what people consider a pathway is extremely ephemeral and
context dependent.
GO:0014069 postsynaptic density
The definition in GO defines the PDS as the cytoplasmic part of the high
electron density, and therefore does not include the membrane part. But if
you look at wikipedia, the definition of the PSD is the high electron
density, also encompasses the transmembrane proteins. The choice of GO was
made after consultation of expert who express what they thought about it.
(for the ones who care, the wikipedia page is very poor and GO represents
the consensus among people working on the PSD. But our definition changes
several times a day, depending if we talk about proteomics or imaging
experiments ...)
I chose GO, but we could take the anatomy ontologies, the sequence
ontology, or any other ontologies dealing with biological content.
For the non-biological content ontologies, such as PATO, unit ontology
etc., most if not all the terms are concepts, things that cannot be
measured. And I have no problem with that.
--
Nicolas LE NOVERE, Computational Neurobiology, EMBL-EBI, Wellcome-Trust
Genome Campus, Hinxton CB101SD UK, Mob:+447833147074, Tel:+441223494521
Fax:468,Skype:n.lenovere,AIM:nlenovere,MSN:nlen...@hotmail.com(NOT email)
http://www.ebi.ac.uk/~lenov/, http://www.ebi.ac.uk/compneur/, @lenovere
------------------------------------------------------------------------------
Gary Merrill
One of the issues swirling around in this discussion is the underlying
question of whether science, and OBO ontologies, should be about "things in
reality" or "conceptualizations". The realist thesis under consideration is
that it should be about only things in reality and not conceptualizations --
and that there is a clear line to be drawn between the two (and that we
have, or can have, criteria for detecting this line in specific cases).
Let me recommend, to those who are unfamiliar with it, a brief paper that
focuses on this issue and may further inform your thoughts about it. This
is:
R. Schwartz. "Starting from Scratch: Making Worlds". Erkenntis, 52 (2000),
pp 151-159.
The paper is only nine pages long and is easily understandable to anyone who
is not steeped in philosophical tradition, lore, and jargon. The examples
it provides are drawn from the domains of astronomy and cosmology, but I
believe you will see direct analogs to biomedical domains such as genetics,
diseases and medical conditions, drugs, etc.
--
View this message in context: http://obo-discuss.2851485.n2.nabble.com/Ontological-Realism-and-OBO-Foundry-Criteria-tp5293729p5305908.html
Sent from the OBO Discuss mailing list archive at Nabble.com.
------------------------------------------------------------------------------
Michel_Dumontier
I think this is exactly what we want - that our assumptions are formally represented such that we discover classes that are inconsistent with our current scientific understanding. Inconsistent classes are those that have no instances, and this is exactly the case for an "impossible" molecule.
m.
-----Original Message-----
From: Janna Hastings [mailto:hast...@ebi.ac.uk]
Sent: Saturday, July 17, 2010 7:14 AM
To: obo-d...@lists.sourceforge.net
Cc: le...@ebi.ac.uk; obo-d...@lists.sourceforge.net
Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry Criteria
Barry Smith
the heading 'inconsistent', then you will immediately obsolete this
term from your ontology, right? If yes, then your view corresponds
exactly to the view I proposed in my response to your FOIS paper.
BS
Barry Smith
>These examples are somewhat helpful. I'd like to see nearer misses.
>Things that are controversial. No one in biomedical informatics
>would plan on including unicorns in their ontology. Where is the
>principle actually helpful in resolving controversy?
We will be presenting an answer to this question in our response to
Merrill's paper, very soon.
>What about encoding negation? If I do an experiment and find that
>E. coli cannot utilize (say) maltose as a carbon source, can I
>capture that in an OBO ontology? It's an important scientific
>fact to capture, but I can't tell whether it would correspond to
>reality in the OBO sense.
This is easy:
http://ontology.buffalo.edu/medo/NegativeFindings.pdf
Incidentally, you should not assume that views expressed concerning
universals, instantiation, realism, etc., on this list are entirely
accurate representations either of what Ceusters and I hold (as we
shall point out, again, in the response to Merrill), or of what OBO
Foundry requires.
BS
Adam M. Goldstein
On Jul 17, 2010, at 1:07, Peter Karp <pk...@AI.SRI.COM> wrote:
> These examples are somewhat helpful. I'd like to see nearer misses.
> Things that are controversial. No one in biomedical informatics
> would plan on including unicorns in their ontology. Where is the
> principle actually helpful in resolving controversy?
>
The unicorn example is meant to make the discussion easier by taking something we are familiar with that is not impossible, but is known not to exist. There are much stranger things than unicorns in science, I think.
But the unicorn example is just that---an example. Many of the theories other efforts on issues about vacuous reference arose in the context of mathematics. Before the solution to Fermat's last theorem was known, people would say things like "well, the solution to F's last theorem is going to require advances in such-and-such disciplines." People like Russell wanted to avoid the conclusion that all the talk like this about the theorem was meaningless, even if the theorem had been a non-theorem. The strategy was to reconstruct discourse about these kind of things in terms that are unobjectionable if the supposed object does not exist.
There are many other unicorn-like examples that come from science. There is a period during which something is posited; it may or may not exist; if it is determined not to exist, there is still some way to speak if it.
I am not saying that this decides the issue about how these things should figure into the BFO. The point is that the problems about absent or non-existent entities cannot be dismissed on the grounds that no one ever seriously discusses unicorns. Moreover the fact is that there really is a need to handle fictions and fantasies because they play a central role in psychiatry, cognitive science, and the like. The FDA insert that comes with a prescription for a schizophrenia drug will say that the drug is intended to reduce the intensity and frequency of hallucinations.
------------------
Adam M. Goldstein PhD, MSLIS
--
z_calif...@shiftingbalance.org
http://www.shiftingbalance.org
http://www.twitter.com/shiftingbalance
--
http://www.itis.gov/servlet/SingleRpt/SingleRpt?search_topic=TSN&search_value=180621
--
(914) 637-2717 (msg)
--
Dept of Philosophy
Iona College
715 North Avenue
New Rochelle NY 10801
Michel_Dumontier
Note the key distinction between intended inconsistencies that arise from *scientific knowledge* and those arising from *errors* in the representation of knowledge. Clearly the latter is of value during the development of a formal ontology using an automated reasoner. As for the former, David pointed out some terms that have become "obsolete" in his ontology, but the rationale "does not exist" is insufficient, and goes to some of what Gary argues - the criterion are poorly expressed. Are the terms simply not used in discourse? do they refer to things that do not follow from our scientific understanding (e.g. at least one of his examples show this - "adult abdominal spiracle 8" -> There are only 7 pairs of spiracles in the adult abdomen )? or do they refer to a logical impossibility (e.g. intersection of two disjoint types) given the current scientific understanding such that they are never instantiable? As an aside, where is the rationale for introducing these terms in the first place?
We should admit that our ontology engineering practices (design & assessment) are either poor, confusing or non-existent. As a community, we should certainly strive to improve and standardize the methodology and practices on clear and sound criteria. I appreciate the time and effort that have gone into building up the OBO Foundry, clearly outlined in Suzanna's email. I appreciate the effort that Barry and colleagues have made (which has influenced my work to a large extent). Many people see the OBO Foundry as a major vehicle towards the improvement of OBO ontologies, and generally in the area of formal ontology. Clearly, significant positional and organizational changes *must* be made to acknowledge and support the wider requirements of the community, otherwise I suspect the pursuit of alternative social structures is inevitable. I think many, including myself, would be willing to meaningfully participate in OBO Foundry's transformation should the opportunity be presented.
m.
[1] http://dumontierlab.com/pdf/2010_BIOONT_hyque.pdf
Richard H. Scheuermann, Ph.D.
Having said this, the Cell Ontology needed to be able to include specific absences in the logical definitions of many hematopoietic cell types (see example below and a detailed description in Masci, AM et al. (2009) “An improved ontological representation of dendritic cells as a paradigm for all cell types” BMC Bioinformatics, 10: 70). The bottom line is that there really are instances of "IgD immunoglobulin complexes" and "maltose utilizations as carbon source", just not in this cell type or the E. coli class of organisms, respectively.
[Term]
id: CL:xxxxxxx
name: IgG-positive double negative memory B cell
namespace: cell
def: "A memory B cell with the phenotype IgD-negative, IgG-positive, CD24-negative, CD38-negative, and CD27-negative." [PMID: 20123131]
is_a: CL:0000981 ! double negative memory B cell
intersection_of: CL:0000981 ! double negative memory B cell
intersection_of: lacks_plasma_membrane_part GO:0071738 ! IgD immunoglobulin complex
intersection_of: lacks_plasma_membrane_part PRO:000001963 ! tumor necrosis factor receptor superfamily member 7
intersection_of: lacks_plasma_membrane_part PRO:000001932 ! signal transducer CD24
intersection_of: lacks_plasma_membrane_part PRO:000001408 ! ADP-ribosyl cyclase 1
intersection_of: has_plasma_membrane_part GO:0071735 ! IgG immunoglobulin complex
Richard
------------------------------------------------------------------------------------------------
Richard H. Scheuermann, Ph.D.
Chief, Division of Biomedical Informatics
John H. Childers Professorship in Pathology
Department of Pathology
U.T. Southwestern Medical Center
5323 Harry Hines Blvd.
Dallas, TX 75390-9072
phone: 214-648-4115
FAX: 214-648-4070
email: richard.s...@utsouthwestern.edu
http://pathcuric1.swmed.edu/Research/scheuermann.html
Robert Hoehndorf
Hi,
> Let us imagine that we concede that empty types are useful to formulate in
> purely logical terms, making no appeal to reality or the existence of
> instances to guide our ontology development. Surely, even so, our
> aspiration in chemical ontology is to encode, in the ontology, precisely
> the physical rules of the feasibility of molecules as we discover them.
I would doubt that. Investigating the physical laws of nature is a
subject of science, not ontology (but then, we may have to define our
understanding of "ontology", see below).
> Encoding structural features as necessary and sufficient conditions on
> classes is an active area of research in chemical ontology at the moment.
Yes, but encoding structural features has nothing to do with physical
rules of feasibility. In an ontology of chemistry, you may have
classes named "molecule", "oxide" and "carbondioxide", and you specify
the features of these classes. Based on these features, you can derive
is-a relations between these classes, because CO2 has an O as part,
CO2 satisfies the conditions for being a molecule, etc. Yet, there is
nothing scientific in these assertions of the ontology; otherwise, you
would have to be able to come up with some kind of experiment that
makes them false, i.e., to experimentally show that something that as
a C and two Os as part is not a kind of thing that has an O as part,
etc.
> If we assume for the sake of argument that we could do this properly, and
> encode the structural features of feasible molecules as necessary and
> sufficient conditions on the class 'molecule', and also encode precisely
> the structural features of this impossible alkane, then a reasoner would
> be able to pick out the class as inconsistent if it is asserted as a
> subclass of molecule. Surely we do not want to encourage a future
Indeed. But then you would have built a universal chemical prediction
tool (at least with respect to one particular scientific theory of
chemistry), and not an ontology. This would be a quite useful encoding
of one scientific theory, but it could not support science; there
would be no progress, because the vocabulary that is defined in this
"ontology" would already entail is empirical validity. Science works
by observation, creating theories, making predictions and testing
these predictions through experiments. What would happen if I
synthesize an impossible molecule (with respect to one theory, that
you used to encode your "Molecule" class in the universal prediction
tool)? I would not have found anything about "Molecule" (which, by its
definition excludes my synthesized chemical), but something else (in
your ontology, I would have found an instance of the class of
impossible things). Maybe now you would revise your "ontology" and
include my newly synthesized structure. But how would you know that
what I have synthesized is a "molecule"? It is /this/ what an ontology
must encode if it is to be useful for data integration and
interoperability in science.
Again, we may have a disagreement about what we mean by
"ontology". You seem to refer to the encoding of one scientific theory
as "ontology". This is okay. But then, to solve the problems that
ontologies like the GO try to solve, we would still need something
else, in which the basic categories that science investigates are
clearly specified, categories about which we can discover new things
-- true or false -- through science.
James Malone
classes corresponding to instances in real life is insufficient. Perhaps
instead, as scientists, we should consider the criteria to be more based
on evidence, after all this is what most science is based on; taking
observations, measurements etc and formulating theory. I think the de
facto OBO Foundry criterion at the moment is that sufficient evidence has
been collected by the community for these instances to exist and that is
what the Realism criteria corresponds to. However, for more hypothetical
cases such as Higgs boson, these classes would fall out of scope. However,
if the criteria were changed to suggest that classes without currently
known instances were included on the proviso that this is flagged
(logicall or otherwise), perhaps by linking to hypothetical evidence,
formula and so on, then this would cover all cases. It would also allow
for the generation of a set of classes that 'only exist in the real world'
(as presently evidenced) something akin to the present OBO Foundry set,
whilst allowing for a wider set of ontologies which included useful, more
hypothetical classes, to be included.
Cheers,
James
--
European Bioinformatics Institute,
Wellcome Trust Genome Campus,
Hinxton,
Cambridge, CB10 1SD,
United Kingdom
Tel: + 44 (0) 1223 494 676
Fax: + 44 (0) 1223 492 468
Philippe Rocca-Serra
Hi James,
To some extent, several resources rely on metadata to somewhat track
such evidence of existence (source and "example of usage" metadata tags
for OBI). Providing a pubmed id means that classes are anchored in
reality and there are documentated experiments supporting defining
classes as done in the ontology.
I believe from the discussion that one of the idea which was floated
around was to allow hypothesis to appear as "specifications", where one
would describe the constraints such an new entity should have.
As you suggest, the identification of evidence satisfying those
requirements would allow promoting hypothezised entities to independent
continuants. We would have some kind of mechanism to automatically
promote or invalidate those specifications.
This still leaves us with the problem outlined by Colin, namely the size
of the field of possibilities. How to decide why hypothesis are
reasonable and should be included (existence of protein as pathogen
agent, viral agent as a cause of AIDS, tectonic plates moving around)
from the rest (memory of water) ? I presume again, we could rely on
consensus and ask domain experts who are working at the leading edge
(even if those disagree!).
Of course, one can deride this argument by stating that we might end up
having creationist views to be represented but I also believe that this
is not what is being discussed in the thread and people want to discuss
'reasonable possibilities' in the field of science:)
cheers
Philippe
--
/-----------------------------------------------\
Philippe Rocca-Serra (PhD)
EMBL-EBI
NET-Project Technical Coordinator
Tel: +44 (0)1223 492553
Fax: +44 (0)1223 494
URL:
http://www.ebi.ac.uk/net-project/
http://isatab.sourceforge.net/index.html
\-----------------------------------------------/
Barry Smith
Foundry distinction between reference ontologies (for settled
science) and application ontologies (for other matters). It is worked
out formally in the referent tracking framework produced by Ceusters
and myself. There we have been concentrating on instance data, rather
than ontologies, but the issue (of how we deal with reasoning about
hypotheticals) pertains just as much to instances as to general
categories. What we still need for the ontology case, as James points
out, is an adequate set of evidence codes for assertions in
application ontologies involving the use of provisional terms for
which instances have not been identified.
BS
Janna Hastings
The criteria for what makes a valid molecule are not determined by the logical definition in an ontology. They are determined by the laws of physics which exist in the reality we try to learn about through science. I do not think your question <how would you know if it is a valid "molecule''> makes any sense at all: if you can successfully synthesise it, then it is a valid molecule. You have not found an instance of an impossible class, you have created an instance of a possible class that was previously believed to be impossible.
In our ontologies, we try to formulate the best current shared community understanding of scientific laws as we can. If it so happens that a molecule that was previously thought to be impossible is successfully synthesised, we have thereby learned something new about the laws of physics governing the feasibility of molecules, and therefore we can immediately change the definition we capture in our ontology in order to reflect our new up-to-date understanding and render the not-impossible molecule no longer inconsistent in our ontology.
That is what is meant by the continuous evolution of ontologies to reflect progress in science.
Cheers, Janna
> -----Original Message-----
> From: Robert Hoehndorf [mailto:leec...@leechuck.de]
> Sent: 18 July 2010 10:25
> To: obo-d...@lists.sourceforge.net
> Subject: Re: [Obo-discuss] Ontological Realism and OBO Foundry Criteria
>
> On Sat, 2010-07-17 at 12:14 +0100, Janna Hastings wrote:
Robert Stevens
for example, for individual glucose molecules. What evidence do we
have for the other entity -- the glucose universal? If we have
another universal of "hexose sugar" what is the evidence that this
entity, as a universal, exists outside a chemist's conception of
categories of chemicals? The thing I've always struggled with is
knowing the difference between a universal and "just how a scientist
talks about something". What do I record as evidence that the entity
hexose sugar exists outside my conception?
Robert.
Barry Smith
>On Jul 17, 2010, at 1:07, Peter Karp <pk...@AI.SRI.COM> wrote:
>
> > These examples are somewhat helpful. I'd like to see nearer misses.
> > Things that are controversial. No one in biomedical informatics
> > would plan on including unicorns in their ontology. Where is the
> > principle actually helpful in resolving controversy?
> >
>
>The unicorn example is meant to make the discussion easier by taking
>something we are familiar with that is not impossible, but is known
>not to exist. There are much stranger things than unicorns in science, I think.
>
>But the unicorn example is just that---an example. Many of the
>theories other efforts on issues about vacuous reference arose in
>the context of mathematics. Before the solution to Fermat's last
>theorem was known, people would say things like "well, the solution
>to F's last theorem is going to require advances in such-and-such
>disciplines." People like Russell wanted to avoid the conclusion
>that all the talk like this about the theorem was meaningless, even
>if the theorem had been a non-theorem. The strategy was to
>reconstruct discourse about these kind of things in terms that are
>unobjectionable if the supposed object does not exist.
>
>There are many other unicorn-like examples that come from science.
>There is a period during which something is posited; it may or may
>not exist; if it is determined not to exist, there is still some way
>to speak if it.
Not quite -- for there is, by hypothesis, no it to speak of.
Certainly we still have the terms, which would, I hope, as soon as
they are determined not to correspond to anything on the side of
reality, immediately be obsoleted in a sensible ontology. But even
then we certainly still have the terms.
>I am not saying that this decides the issue about how these things
>should figure into the BFO. The point is that the problems about
>absent or non-existent entities cannot be dismissed on the grounds
>that no one ever seriously discusses unicorns. Moreover the fact is
>that there really is a need to handle fictions and fantasies because
>they play a central role in psychiatry, cognitive science, and the
>like. The FDA insert that comes with a prescription for a
>schizophrenia drug will say that the drug is intended to reduce the
>intensity and frequency of hallucinations.
Bear in mind that allucinations and fantasies certainly exist, as do
prescriptions, schizophrenia, drugs, etc.
BS
Robinson, Peter
I cannot resist making the statement that the issue of "Ontological Realism and OBO Foundry Criteria" is not of great importance for users of ontologies. There are two reasons for this statement.
1) Annotation Level:
Many of the controversial topics in biomedical research can be fully captured using existing ontologies. For instance, to describe a unicorn, we can use all the terms used to describe a horse as well as the term for horn from the appropriate anatomy ontology and use the term for flight that is used to describe birds...what I mean is that many controversies are of the sort, "should protein X be annotated by GO term Y?" For this situation there is no need for new terms, the question is whether the annotation is correct, and this can be addressed using evidence codes.
2) Term Level:
If on the other hand a research groups posits the existence of some novel entity that is received controversially by the community, the only thing that is missing is an identifier for the new entity, which for a start can be chosen by that group. Again the proposed characteristics of that entity can at least ideally be described using the "realist" terms of existing ontologies. Presumably, as soon as a second research group begins to publish about the new entity, they will have confirmed the initially controversial but now less so observations of the the first group, and the new entity will be welcomed as a Term into the appropriate ontology. As long as there is only one group, the need for communication about the new entity is presumably not pressing.
Therefore, I would tend to support the view that ontologies should take a realist stance and not include terms about things which have not been proven or at least made plausible by scientific research. The disadvantages of having terms whose relevance and correctness are uncertain outweighs the advantages of those few cases in which uncertain hypotheses need new ontology terms to be communicated.
Best wishes Peter
PD Dr. med. Peter N. Robinson, MSc.
Institut für Medizinische Genetik
Charité - Universitätsmedizin Berlin
Augustenburger Platz 1
13353 Berlin
Germany
+4930 450566042
peter.r...@charite.de
http://compbio.charite.de
http://www.human-phenotype-ontology.org
Stefan Schulz
>
> Bear in mind that allucinations and fantasies certainly exist, as do
> prescriptions, schizophrenia, drugs, etc.
Hallucinations and fantasies are "about" something. How to refer to
this "something" if it has no instances in reality? As Peter R.
pointed out, it is possible to refer to something "non-existing" by
combination. See attached a representation of different unicorn
hallucinations constructed exclusively by classes that have instances.
Observe how the different kinds of Hallucinations classify as expected
(e.g. FlyingWhiteUnicornHallucination is classified under
WhiteUnicornHallucination). However, as soon as we axiomatically state
that there are no unicorns, e.g. by adding Horse subClassOf hasPart
only (not Horn), then the different kinds of unicorn hallucinations
are no longer distinguishable, which would certainly reduce the value
of such an ontology e.g. for the annotation of delusional ideas
expressed by schizophrenic patients.
Cheers,
Stefan
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
> _______________________________________________
> Obo-discuss mailing list
> Obo-d...@lists.sourceforge.net
> https://lists.sourceforge.net/lists/listinfo/obo-discuss
>
--
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
Barry Smith
>we can find lots of evidence for instances. We have lots of evidence,
>for example, for individual glucose molecules. What evidence do we
>have for the other entity -- the glucose universal?
The fact that scientists describe multiple instances thus.
Repeatedly. Effectively. Usefully. (Also, of course, revisably --
scientist might teach us better.)
In addition, of course, there is the chemist's conception of the
corresponding category.
So scientists are able to describe those repeated instances
effectively because they are somehow able to match their conception
to something in those instances, something repeatedly observed. This
cannot BE the scientists' conception, because it was there before the
scientists came along.
BS
Michel_Dumontier
m.
-----Original Message-----
From: Barry Smith [mailto:phis...@buffalo.edu]
Michel_Dumontier
You seem to imply that all the terms we require are already present, but we know that not to be true, as curation is an ongoing activity precisely because new terms are coined by scientists as part of their scientific discourse. Importantly, the new terms refer to entities with different characteristics than have been previously described. They are used to formulate hypotheses about the world. This is *crucially* important in scientific communication and has implications for the experimental validation of scientific theories. I fully disagree that it is 'not pressing'.
Moreover, we should not restrict the value of ontologies to what we are currently doing with them, but rather, we should think about what we *could* be doing with them. I've argued in this thread that there are use cases that are currently unsatisfied and that this hinders the development of newly emerging approaches. I've also argued that we should investigate methods that bridge the gap between ontologies that contain discourse and those with only instantiable types, so as to satisfy those particular needs/desires that you and others have expressed. We should welcome a diverse set of needs so as to facilitate the use ontologies across a number of scientific endeavors.
m.
-----Original Message-----
From: Robinson, Peter [mailto:peter.r...@charite.de]
Sent: Sunday, July 18, 2010 8:26 AM
To: obo-d...@lists.sourceforge.net
Robert Hoehndorf
Hi,
> The criteria for what makes a valid molecule are not determined by the
> logical definition in an ontology. They are determined by the laws of
> physics which exist in the reality we try to learn about through
> science. I do not think your question <how would you know if it is a
There are many laws. How do you determine which are about molecules?
Before you can even start to investigate molecules, you need to know
what it means to be a molecule -- this is what ontology can provide
for you. There is no science in this, nothing that can be falsified by
experiments. There is a kind of thing you find useful to formulate
your theory, you call it "molecule".
> synthesise it, then it is a valid molecule. You have not found an
> instance of an impossible class, you have created an instance of a
> possible class that was previously believed to be impossible.
Only in your conception, it would not be a molecule, because
possibility and impossibility is already encoded in the meaning of the
term!
> In our ontologies, we try to formulate the best current shared
> community understanding of scientific laws as we can. If it so happens
I disagree. There is nothing scientific about "apoptosis is-a cell
death" or "H2O is-a molecule", and neither should there be. There is
nothing that can be made false, nothing that can be predicted from
that alone. There are two kinds of entities, kinds of things science
is interested in, H2O and molecule. Once we know what it means to be a
molecule and what it means to be H2O (they are just labels, btw.,
everything about them is in the definition) can we investigate and
state laws both about molecule and H2O.
> that a molecule that was previously thought to be impossible is
> successfully synthesised, we have thereby learned something new about
> the laws of physics governing the feasibility of molecules, and
> therefore we can immediately change the definition we capture in our
> ontology in order to reflect our new up-to-date understanding and
> render the not-impossible molecule no longer inconsistent in our
> ontology.
You do not answer /how/ you would know that you have found out
anything about a molecule! In your "ontology", the "molecule" class
was defined to exclude the impossible molecule, so whatever you have
synthesized is not a molecule in your ontology. So there must be some
kind of meta-molecule class by which you can identify kinds of
molecules, but this is not part of your ontology. Yet, as you seem to
agree, this is what is needed to do science.
Rob.
> That is what is meant by the continuous evolution of ontologies to
> reflect progress in science.
>
>
>
> ------------------------------------------------------------------------------
> This SF.net email is sponsored by Sprint
> What will you do first with EVO, the first 4G phone?
> Visit sprint.com/first -- http://p.sf.net/sfu/sprint-com-first
> _______________________________________________ Obo-discuss mailing list Obo-d...@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/obo-discuss
------------------------------------------------------------------------------