Add unit, measurement, quantitative values

[jamesaoverton]

Related issue on PATO tracker: https://github.com/pato-ontology/pato/issues/101

[cmungall]

James' proposal: https://docs.google.com/document/d/14qqp0M2dgefDFMvB4mmwpxrhoo4UGwd1KLZcoWCcqss/edit

[cmungall]

I discovered a related effort, CDT:

https://ci.mines-stetienne.fr/lindt/v2/custom_datatypes

It provides datatypes for what we would call qualities/characteristics,

E.g.

https://ci.mines-stetienne.fr/lindt/v3/custom_datatypes#length

"1 km"^^cdt:length
"1.609344E+6 mm"^^cdt:length
"5280 [ft_i]"^^cdt:length

I'm not totally sure I grok any advantages to this system, over the proposed COB system where we would instead write

"1"^^somePrefix:km

This is being discussed here:

https://lists.w3.org/Archives/Public/semantic-web/2020Jul/0157.html

UPDATE

Eric proposes a scheme identical to ours:

https://lists.w3.org/Archives/Public/semantic-web/2020Jul/0169.html

With a defense of the scheme here: https://lists.w3.org/Archives/Public/semantic-web/2020Jul/0174.html

PPS objects: https://lists.w3.org/Archives/Public/semantic-web/2020Jul/0175.html

Antoine also objects: https://lists.w3.org/Archives/Public/semantic-web/2020Jul/0178.html

But I am not sure I understand/buy the objeection. There is no need to parse, any more than there is a need to parse "en" in a language literal

The thread also talks about restrictive licensing on UCUM. This would be an issue for us, especially if we need new units specific to biology, we want an open resource on GH... but how much of UO is not in UCUM?

[mellybelly]

probably worth considering LOINC interop: https://loinc.org/usage/units/ and/or various units value sets: https://www.hl7.org/fhir/valueset-ucum-units.html https://cdebrowser.nci.nih.gov/cdebrowserClient/cdeBrowser.html#/dataElement

@DaveraGabriel has been doing a lot of QC and transformation on units and may have some perspective

[kaiiam]

If you guys end up deciding to go with a well flushed out version of UO, I will have scope to spend a good amount of time on it.

[cmungall]

I believe we are centering around a different representation, no longer proposing unit datatypes, instead something like

[
 a pato:Weight ;
 characteristic-of person0001 ;
 has-measurement [
    unit: UOM:m
    value: nnnn
 ]
]

[LarsVogt]

Hi all, Chris pointed me to this discussion. If the pattern from above should follow a general and expandable pattern of how to model measurement data (instead of having several more specific patterns), I have two questions or comments:

1) A given quality can have more than one measurement, because measurements are the results of processes and use methods, devices, and sometimes even models, which can influence the result (in addition to the agent). Therefore, something like a measurement datum resource is needed, to which you can connect the value and unit of the measurement result, so that a given quality can have several measurement data. 2) A measurement datum can comprise more than one value (e.g., a confidence interval). Thus, in a general pattern, we also need the value specification resource, I think.

In combination, it allows the specification of e.g. a measurement with 95% confidence interval or linking the numerator and denominator of a ratio value to its underlying basic measurement data.

[ddooley]

A has-measurement model would be great. I agree with above, that a characteristic could be measured at least at several points in time, potentially with its own accuracy and precision. IAO itself ran into that complexity by realizing each point in time was itself a measurement. OWL itself gets complicated with any hint of time, but here we're dealing with the desire to have ontology directly model graph data - to provide validated syntax for the data statements, without utilizing reasoning power - which is a fine use of ontology to bring order to the world.

What I'd suggest in a new thread is development of observational data structures that go beyond just simple 1 dimensional measures. We could have a measurement/observation data structure which is a measurement set made at a certain time t, and it only contains an entity's qualities and their units and values. Another example, a longitudinal observational data structure of an entity quality, made with the same unit, but at different times t. In theory all of these could normalize to a big set of characteristic + time/location/altitude/other dimension + unit + precision/accuracy entities.

Also, what is essential is that a measurement can be categorical, i.e. a value, no unit (e.g. hot / cold). A measurement can be ordinal, so numeric but not necessarily with even intervals. A measurement may be scalar, i.e. numeric, but that doesn't necessitate a unit either. And finally can we define measurements without the assumption of real world? I'd like to run a simulation and not have to use different names for these measurement entities just because they measure characteristics of simulated entities.

[ddooley]

@jamesaoverton do you want to add your 'has characteristic' and 'measurement datum' examples that you showed in COB workshop here?

[jamesaoverton]

My plan is to write up something longer and link to it from here.

[ddooley]

@jamesaoverton update on your document for COB?

I was going to propose an RO "[characteristic] 'has measure' some [characteristic | data item]", and inverse "is measure of", that would support and simplify a QQV model. In allowing both characteristic and data item in its range, it covers both categorical and numeric / scalar / unit bearing measures of characteristics.