question re: linking databases to ontologies

[Jessica Turner]

Hi folks--We're starting the process of annotating some of our imaging databases with CogPO terms, to make sure we have the whole system worked out for at least a few examples.  I want to make sure I'm thinking about it correctly, so let me run this by you before I go off creating thousands of instances/concepts.

 

We have a concept in CogPO called the Oddball Paradigm; I have two datasets in my database that were collected with different specific oddball paradigms. I am planning to create two instances of the Oddball class, call it Exp1 and Exp2. I would put the URI for Exp1 in the database field that's right for that dataset, and the URI for Exp2 in the right field in the database.  (Do individuals get URIs, or only classes? If it's only classes then all of these things become classes and subclasses, which is fine by me but a little silly.)

 

I was thinking of the information about that Exp1 instance (what experimental conditions it specifically has, the stimuli, response type, etc.) would be in my CogPO owl file, as "individuals" in the ontology. That's the point of the ontology, to describe information about the paradigm that isn't in the database explicitly. But I'm thinking of it more as a knowledge base, with individual terms for all the experiments and conditions that we need, that use the CogPO terms.   There will be *thousands* of these experiments and conditions, many for every paper published, but that's ok, no one will ever need to know them all.  They'll be queried based on information about them (what stimuli they used, etc.) and not by label.

 

The reason I'm doing it that way is because I want to be able to go back later and add more information about the specific paradigm--to be able to say that Exp1 uses a 1000 Hz tone while Exp2 uses white noise as a stimulus, for example--that I don't need to model right now, but someone might need to model later.  If I simply annotate it by the concept of "Oddball paradigm" as a general class, then I don't have the information that I need about the experiment; I also lose the granularity I want. The Exp1 paradigm will never change--the data are collected, it is what it is; whether it actually is an oddball paradigm (or whether oddball paradigms as a class continue to be considered such) might change as future thinking about it changes. I'm trying to build a system that doesn't explicitly contain all the information now, but sets up a framework whereby needed info/characteristics about the experiments could be added later.

Does this sound reasonable to you?  Is this at all how it's being done for other databases and domains? Should I just try it and see how it goes?

 

Thanks,

Jess
--
Jessica Turner, Ph.D.
Associate Professor, Translational Neuroscience
Mind Research Network
1101 Yale Blvd. NE
Albuquerque, New Mexico 87106
New Ph: 505-272-2869
email: jtu...@mrn.org

[Finn Aarup Nielsen]

Hi Jess,

In Brede I do not create a class for each paradigm in a paper. Instead I
just tags each experiment to a class on a "reasonable" level in the
paradigm/cognitive component hierarchy and extend the hierarchy "as I go
along". It seems to me that almost all functional neuroimaging experiment
each apply their own variation of a paradigm. You would never have a pure
(say) "Iowa Gambling Task" there will be endless variations on the timing,
stimulus, responses (left hand/right hand, thumb, joystick). In
event-related fMRI experiment you also have the ability to make different
results based on the timing of the constrast, e.g., "Anticipation phase"
and "Experience phase"
(http://neuro.imm.dtu.dk/wiki/Brain_maps_of_Iowa_gambling_task)

I think it will be good to be able to extend the description of an
experiment.

/Finn

___________________________________________________________________

Finn Aarup Nielsen, DTU Informatics, Denmark
Lundbeck Foundation Center for Integrated Molecular Brain Imaging
http://www.imm.dtu.dk/~fn/ http://nru.dk/staff/fnielsen/
___________________________________________________________________

[Gwen Frishkoff]

Hi Jess! We just presented an approach to data annotation in RDF using
NEMO ontologies at an NIH/NIF-sponsored workshop. We are pretty pleased
with the results, and I'd be glad to review this with you and Angie if
you like and discuss how it might or might not be applicable for your
fMRI expt data.

Thanks,
Gwen

Gwen Frishkoff, Ph.D.
Assistant Professor, Department of Psychology
Associate Member, Neuroscience Institute
140 Decatur Street
Urban Life, Rm 712
Georgia State University
Atlanta, Georgia 30303-3083

Email: gfris...@gsu.edu
Phone: 404-413-6303
Fax: 404-413-6218

"The price of training is always a certain 'trained incapacity': the
more we know how to do something, the harder it is to learn to do it
differently."
-- A. Kaplan (The Scientific Method in Behavioral Science)

[anita bandrowski]

Hi Jess and Finn,

I am struggling with a similar problem, as I attempt to translate our resources (ex., BREDE) into the ontology and I am trying to figure out if they should be instances of Resource.  It seems wrong to define these experiments or resources as classes in the direct 'is a' tree, but I am not completely sure why.

As Jess mentioned, any one experiment could exist whether or not it is associated with a specific paradigm (Oddball), so it seems that the 'is a' tree is not useful for this purpose.  In handling resources, I was considering using the 'has role' relationship, as any one resource can have role database even if the database class today includes things xml datasets that may not have included a couple of years ago.

Any thoughts on this??
Best,
anita

--
You received this message because you are subscribed to the Google Groups "neurolex" group.
To post to this group, send email to neur...@googlegroups.com.
To unsubscribe from this group, send email to neurolex+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/neurolex?hl=en.

--
Anita Bandrowski, Ph.D.
NIF Project Curator
UCSD 858-822-3629
http://neuinfo.org
9500 Gillman Dr.#0446
la Jolla, CA 92093-0446

[Finn Aarup Nielsen]

Hi Anita,

I guess I based on the BrainMap system use an experiment that 'has a'
paradigm or 'has a' cognitive component. One may want to have two 'is a'
systems: One for cognitive components another for 'paradigms'/tasks. In
the hierarchy in the Brede database
http://neuro.imm.dtu.dk/services/brededatabase/WOEXT_1.html I really do
not distinguish between the two, since I sometimes find it hard.

/Finn

>> http://www.imm.dtu.dk/~fn/ <http://www.imm.dtu.dk/%7Efn/>

>> http://nru.dk/staff/fnielsen/
>> ___________________________________________________________________
>>
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "neurolex" group.
>> To post to this group, send email to neur...@googlegroups.com.
>> To unsubscribe from this group, send email to

>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>

[Russ Poldrack]

I think the idea is that the cognitive atlas and cogpo will link together to provide the links between tasks (which cogpo represents) and  cognitive functions (which the cognitive atlas represents). 

cheers

rp

--
Russell A. Poldrack
Director, Imaging Research Center
Professor of Psychology and Neurobiology
University of Texas at Austin
3925-B W. Braker Ln.
Austin, Texas 78759

Phone: 512.232.9504
Fax:512.232.4202

Email: pold...@mail.utexas.edu
Web: http://www.poldracklab.org

[Jessica Turner]

With reasoning from evidence working in there somewhere, but what Russ said is definitely the general idea.

 

But to get back to parallels between Anita's question and mine originally: We're at a point where we are stepping out of the general, abstract ontology and into linking the ontological concepts to real world stuff, and it's the link that is tricky.   I like the idea that the "stuff" should have a URI that never changes; whether its place in the ontology gets moved about and reclassified, it itself shouldn't ever be deprecated unless it itself changes. If I delete the database and start it up again somewhere else I can change the URI for that database; otherwise I want to keep it the same, because it refers to the same thing. (Unlike the URI that identifies a term in the ontology that might be deprecated/ fundamentally redefined or removed/etc.) 

 

At first glance I don't like the "has role" idea--the "stuff" that is the database "plays the role of" a database?  My protocol "plays the role of" an Oddball protocol?  My dog at home "plays the role" of a terrier, if I'm not sure about his breed and may reclassify him as another breed if more info comes to light?  :)  Seems like is_a is what that relationship is about, but I want a special dispensation for instances so that if I end up moving their representation about in the ontology I don't have to change their URI.

 

thanks,

Jess

[Gwen Frishkoff]

Hi Jess.

Jessica Turner wrote:
> With reasoning from evidence working in there somewhere, but what Russ said
> is definitely the general idea.
>

I can show you how we've started to work evidence (RDF data
representations for EEG/ERP patterns) into this process for NEMO.

> But to get back to parallels between Anita's question and mine originally:
> We're at a point where we are stepping out of the general, abstract ontology
> and into linking the ontological concepts to real world stuff, and it's the
> link that is tricky. I like the idea that the "stuff" should have a URI
> that never changes; whether its place in the ontology gets moved about and
> reclassified, it itself shouldn't ever be deprecated unless it itself
> changes. If I delete the database and start it up again somewhere else I can
> change the URI for that database; otherwise I want to keep it the same,
> because it refers to the same thing. (Unlike the URI that identifies a term
> in the ontology that might be deprecated/ fundamentally redefined or
> removed/etc.)

I can review the nuts & bolts of how to do this -- one way -- not nec
the only or best way! -- when we talk on Monday.

>
> At first glance I don't like the "has role" idea--the "stuff" that is the
> database "plays the role of" a database? My protocol "plays the role of" an
> Oddball protocol? My dog at home "plays the role" of a terrier, if I'm not
> sure about his breed and may reclassify him as another breed if more info
> comes to light? :) Seems like is_a is what that relationship is about, but
> I want a special dispensation for instances so that if I end up moving their
> representation about in the ontology I don't have to change their URI.
>

I think you'll resonate to the approach we've taken in NEMO. We are
using the ontologies precisely to do classification of data instances,
which are represented with URI -- but in RDF rather than OWL.

Anyway, it will be interesting to see how our methods might be extended
to handle the data types (cognitive behavioral and fMRI?) that you have
in mind for cogPO.

Gwen

>>>>> http://www.imm.dtu.dk/~fn/ <http://www.imm.dtu.dk/%7Efn/<http://www.imm.dtu.dk/~fn/>>

>>>>>
>>>>>
>>>>> http://nru.dk/staff/fnielsen/
>>>>> ___________________________________________________________________
>>>>>
>>>>>
>>>>> --
>>>>> You received this message because you are subscribed to the Google
>>>>> Groups
>>>>> "neurolex" group.
>>>>> To post to this group, send email to neur...@googlegroups.com.
>>>>> To unsubscribe from this group, send email to
>>>>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>

>>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>>

>> Russell A. Poldrack
>> Director, Imaging Research Center
>> Professor of Psychology and Neurobiology
>> University of Texas at Austin
>> 3925-B W. Braker Ln.
>> Austin, Texas 78759
>>
>> Phone: 512.232.9504
>> Fax:512.232.4202
>>
>> Email: pold...@mail.utexas.edu
>> Web: http://www.poldracklab.org
>>
>>
>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "neurolex" group.
>> To post to this group, send email to neur...@googlegroups.com.
>> To unsubscribe from this group, send email to

[Maryann Martone]

I'm not sure exactly what you mean by this.  You wouldn't be putting in specific values in the class Oddball paradigm, because the 1000 Hz tone only applies to a particular instance.  I'm assuming that you have that information in the database already under that experiment.  Now if you want to turn the entire relational database record into an RDF instance graph, you can.  For CCDB we use a hybrid approach:  field values are annotated with the ontology class URI where appropriate.  If we have a graph model for relating different fields or tables in CCDB, we store those as instances in our Cellular Knowledge Base with pointers back to the ID's in CCDB.  That's the way we do it in theory, anyway.

--

You received this message because you are subscribed to the Google Groups "neurolex" group.
To post to this group, send email to neur...@googlegroups.com.

To unsubscribe from this group, send email to neurolex+u...@googlegroups.com.

[Jessica Turner]

We're actually in a situation where the information is NOT in the database, and that's what makes this a little weird. We are capturing information (somehow--it's part of the knowledge base) that does not always map to a field or table in database.  It's information we know about, it'd be in the paper about the data, but it is not represented explicitly anywhere.  Thus I wanted to add it to a RDF store that technically isn't part of the ontology, which seems to be the consensus--anything that's useful that isn't in the ontology per se goes in a knowledge base like an RDF store. .

 

Thanks,

Jess

[Gwen Frishkoff]

Hi Jess.

Where's the info coming from? Is it from BrainMap? Or from the fBIRN
database? Or from summary data reported in papers that are not in either db?

Or are you trying to represent info that is not reported anywhere (yet),
but might be?

Gwen

Gwen Frishkoff, Ph.D.
Assistant Professor, Department of Psychology
Associate Member, Neuroscience Institute
140 Decatur Street
Urban Life, Rm 712
Georgia State University
Atlanta, Georgia 30303-3083

Email: gfris...@gsu.edu
Phone: 404-413-6303
Fax: 404-413-6218

"The price of training is always a certain 'trained incapacity': the
more we know how to do something, the harder it is to learn to do it
differently."
-- A. Kaplan (The Scientific Method in Behavioral Science)

>>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>.

>>>
>>>
>>> For more options, visit this group at
>>> http://groups.google.com/group/neurolex?hl=en.
>>>
>> --
>> You received this message because you are subscribed to the Google Groups
>> "neurolex" group.
>> To post to this group, send email to neur...@googlegroups.com.
>> To unsubscribe from this group, send email to

>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>.

[Jessica Turner]

Hi Gwen--The info I'm talking about is coming out of the investigator's heads, at the moment. BrainMap is fine, it's set up the way Maryann was thinking, where every table/column maps to a concept in CogPO. It's the databases where the only thing stored was some idiosyncratic string that is meaningful only to the PI that we are dealing with--trying to make that information explicit and useful.

 

Thanks,

Jess

To unsubscribe from this group, send email to neurolex+u...@googlegroups.com.

[Gwen Frishkoff]

Hi. Makes sense. Sounds like you have general concepts in the ontology,
and actual data in RDB, where "actual data" include strings (e.g.,
idiosynchratic condition labels, which presumably map to standardized
labels in the ontology), as well as numbers (like stimulus duration,
rate of presentation, etc.).

So you might have some study-specific label for an expt condition, like
"cell_1," and assert that it is an instance of condition_label category:
e.g.,

<"cell_1"> rdf:type <related_condition_label>

This could be asserted either in the ontology or in the RDF db, but I
think the RDF db might be more extensible than the ontology. I.e., you
can make the RDF db as big as you want, but the ontology might get unwieldy?

You would then declare a link between <related_condition_label> and the
actual experiment category class in your ontology:

<related_condition_label> is_about <related_condition>

You could cut out the middle man, too, and just assert that cell 1 is_a
related_condition. I'm a little worried that could lead to some logical
problems downstream, though, if we're not careful.

My two cents. :)

>>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>>.

>>>>>
>>>>>
>>>>>
>>>>>
>>>>> For more options, visit this group at
>>>>> http://groups.google.com/group/neurolex?hl=en.
>>>>>
>>>>> --
>>>> You received this message because you are subscribed to the Google Groups
>>>> "neurolex" group.
>>>> To post to this group, send email to neur...@googlegroups.com.
>>>> To unsubscribe from this group, send email to
>>>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>

>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>>.

[Jessica Turner]

Yup, that's pretty much the way I'm going about it--and seems to link up with the presentation I heard yesterday about scalable ontologies, too. http://ontolog.cim3.net/cgi-bin/wiki.pl?ConferenceCall_2010_07_22

To unsubscribe from this group, send email to neurolex+u...@googlegroups.com.

For more options, visit this group at http://groups.google.com/group/neurolex?hl=en.

--

[Gwen Frishkoff]

Ah -- thanks for the link!!

>>>>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>
>>>>>>> <neurolex%252Buns...@googlegroups.com<neurolex%25252Bun...@googlegroups.com>>>.

>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>>
>>>>>>> For more options, visit this group at
>>>>>>> http://groups.google.com/group/neurolex?hl=en.
>>>>>>>
>>>>>>> --
>>>>>>>
>>>>>> You received this message because you are subscribed to the Google
>>>>>> Groups
>>>>>> "neurolex" group.
>>>>>> To post to this group, send email to neur...@googlegroups.com.
>>>>>> To unsubscribe from this group, send email to
>>>>>> neurolex+u...@googlegroups.com<neurolex%2Bunsu...@googlegroups.com>
>>>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>

>>>>>> <neurolex%2Bunsu...@googlegroups.com<neurolex%252Buns...@googlegroups.com>
>>>>>> <neurolex%252Buns...@googlegroups.com<neurolex%25252Bun...@googlegroups.com>>>.