core term: variable

[Chris Stoeckert]

Dear Denriers,

Variable is not currently a term in OBI. It is a synonym of variant which is in PATO: 

id: PATO:0001227
name: variant
def: "A variability quality inhering in a bearer by virtue of having or exhibiting variation." [Dictionary:http\://dictionary.reference.com/]
subset: value_slim
synonym: "variable" EXACT []
is_a: PATO:0001303 ! variability

So the proposal is that variable is a quality imported from PATO. This quality could be associated for OBI with the input or output of processes. It could also be restricted in OBI as independent and dependent. 

Note this is also tracker item:  2078188 dependent and independent variables assigned to Alan.

Thoughts?

Chris

[Melanie Courtot]

Hi Chris,

I think that the issue for variable is how to apply it to dependent continuant (namely data), which would relate to the BFO issue of having qualities/roles for continuants. This should be discussed with Barry and his on the agenda, I'm not sure we can proceed with BFO's input on that one.

Personally, I am confused by the use of variables and parameters, and I am unsure what we are usually referring to, so I don't have a clear idea how to proceed on those.

If I set the voltage to 200Volts on my Flow Cytometer, that is a parameter on my instrument but a variable in my experiment. Is that correct?

The data transformation branch published a document at http://docs.google.com/Doc?id=dzprnmw_68cs654hfh&hl=en if that helps.

For me, tiny examples for each case would definitively be helpful :)

Cheers,

Melanie

[Chris Stoeckert]

Hi Melanie,

OK - status of variable is that it is stuck by a limitation of BFO. 

Regarding your example, the voltage setting can be a variable (it has the property to be varied) but for practical purposes - and I would like OBI to be practical ;-) - an experimental variable is evaluated explicitly as part of the experimental design. If you are not monitoring the flow cytometry results as a function of voltage settings, then IMO it's not a variable. Hope this helps. 

Cheers,

Chris

[Melanie Courtot]

Hi Chris,

Thanks for the helpful reply.

So in the case of the Flow Cytometer, the voltage is a parameter of the instrument, and would also be a variable if we were to study the effect of voltage variation on our results for example.

So it seems a variable, in the OBI context, makes sense only at the Plan level (Investigation design). 

Taking an other example that we just discussed with Ryan: we decide to study the effect of a drug dose in treatment of a disease.

My general investigation involves several step, e.g. inject drug to mice, measure size of tumors, compute results.

In the individual steps "injecting drug" and "averaging results" the drug dose is a parameter. The drug dose would be a variable of the general investigation.

Does that mean that data transformations, being individual processes, should never have to deal with variables?

Thanks,

Melanie

---

Mélanie Courtot

TFL- BCCRC

675 West 10th Avenue

Vancouver, BC

V5Z 1L3, Canada

[Chris Stoeckert]

Hi Melanie,

Does that mean that data transformations, being individual processes, should never have to deal with variables?

Hmm, there are a couple issues here. Can individual processes have variables and do we distinguish mathematical variables from experimental variables?

If you perform a gating on the flow cytometry run, you partition event data. Variables associated with the events are for example intensity of transmitted light at different wavelengths (e.g, measure scatter and fluorescein absorption). The wavelengths are parameters. The values of light intensity at those wavelengths in the data set are variables of the gating process. 

So a data transformation makes use of variable properties of data and therefore does have to deal with them I think. The variables of a data transformation are the properties it exploits of the data in order to perform the transformation.

How's that?

Chris 

[Bjoern Peters]

I think we have to be more explicit with what we mean by 'variable:.

I agree with Chris definition of 'experimental variable', which is
primarily what we want to capture in OBI. I agree with Melanie that this
should be part of the 'plan level' of an investigation. 'independent
variable' and 'dependent variable' should therefore be part of the
'study design' specification, which would make them specifications
themselves.

Then there are 'mathematical variables' used in mathematical functions,
which do occur in data transformations.

PATO seems to capture 'presence of variation', and does not seem to be
very usable for us.

- Bjoern

--
Bjoern Peters
Assistant Member
La Jolla Institute for Allergy and Immunology
9420 Athena Circle
La Jolla, CA 92037, USA
Tel: 858/752-6914
Fax: 858/752-6987
http://www.liai.org/pages/faculty-peters

[Frank Gibson]

I have seen this discussion, both on the instrument, dt, relations and
IAO lists with very little end decision. Maybe this should be a key
cross branch discussion at Vancouver?

We need to be able to describe and distinguish (if we believe them to
be different) the following RU's

-parameter
-variable (there are mathematical variables and I like Chris's
definition of an experiment_variable)
-device_setting (at the minute this is very similar to the proposal to
deal with measurements)
-and probably how to represent a measurement is loosely related to
these issues as they potentially all have a value and a unit

I would very much like to have a conclusion on these issues and I
think they would be a very useful core RU discussion at Vancouver

Frank

--
Frank Gibson, PhD
http://peanutbutter.wordpress.com/

[Gully]

Hi everyone,

I haven't been very active on these lists, but have been watching with
much interest. I'm working with Alan Ruttenberg on a knowledge
engineering system that uses OBI and is based on experimental
variables. This issue is therefore central to our formalism.

There are a couple of distinctions that I'd like to ask about:

Can the OBI currently represent these different cases?
(a) independent variables that take multiple values within an
experiment (signifying, for example, the 'control' and 'experimental'
groups)
(b) independent variables that take a single value (for example,
this experiment was performed on 'male adult rats, weighing 300-350
g'). This is important to note, since these values place the
experiment in the context of other experiments, and may be used within
a meta-analysis.
(c) dependent variables that are simply measurements derived from
instruments with no manipulations
(d) derived variables that have been obtained through processing of
other data.
(e) 'proxy' variables where one variable is linked to another via a
vague or non-specific relation (i.e., 'time-spent freezing' is a proxy
for 'fear', which may contribute to the high-level interpretation of a
study).

It sounds as though, you guys are separately dealing with parameters
being used in the protocol and variables being set and measured in the
study design. Is this correct?

Also, the reasoning invoked using the experimental results is
interesting. When you are interpreting data results involving
transformations and data points, you should be able to invoke the use
of variables (but if only parameters can be associated with data, I'm
already confused).

Alan, is there anything you'd like to comment on here?

Thanks in advance for your thoughts.

Gully

[James Malone]

Hi Gully,

I think it is safe to say for the cases you identify OBI can not explicitly represent the concept of 'variable' dependent or otherwise as yet, though these are good use cases you describe.  I (or Melanie, I can't recall now) sent a list of use cases we have from the data transformation branch, some of which were along similar lines to the ones you outline below.  I'm thankful to Chris for also raising this issue again so we can seek a resolution.  I was interested in Bjoern's observation that there appear to be two types of variables we are aiming to capture, the mathematical variables and experimental variables.  Will this discussion be on a call some time?  Certainly it is useful to go over this in the workshop in Vancouer.

Thanks,

James

[Bjoern Peters]

Hi Gully,

This is a good breakdown. I still think we should reduce the use of the
word 'variable' by itself, as everyone seems to have something else in
mind. James is right that we don't yet have any of these in as
'variable' terms, but I believe some may be doable with what we have:

c) 'dependent variable' = data item and is_specified_data_output_of some
assay
synonyms would be 'measurement data', 'primary data'
d) 'derived data' = data item is is_specified_data_output_of some
('data transformation and has_specified_input some measurment data)
e) 'proxy data' = data item and is_about some (continuant is_proxy_for
some continuant)

That leaves a and b:

(a) independent variables that take multiple values within an
experiment (signifying, for example, the 'control' and 'experimental'
groups)

which should be identified as part of the plan how to make the
experiments, e.g. study design or protocol. We had a pretty good
modeling for this on a use case for Jennifer. Need to dig that up.

(b) independent variables that take a single value (for example,
this experiment was performed on 'male adult rats, weighing 300-350
g').

This seems to be simply the description of the experimental process,
which can just be captured in the description of the assays and material
transformations performed. I would not use the term 'variable' here. In
your example, e.g. assay has_specified_input some (rat and has_role some
evaluant_role and has quality weight_range (xyz)). Alan will know best
how to capture absolute values for the weight range.

- Bjoern

> <http://peanutbutter.wordpress.com/>
>
>