I think Allelic Frequency might be an overloaded term.
Perhaps everyone understands exactly what it means, but just in case,
here is how this is defined:

LOINC:
https://r.details.loinc.org/LOINC/81258-6.html?sections=Comprehensive
The allelic frequency is the relative frequency of a particular allele
in the population, expressed as a number from 0 to 1.

FHIR (STU3):
A physical quality which inheres to the allele by virtue of the number
instances of the allele within a population. LOINC Code: 81258-6
https://s.details.loinc.org/LOINC/81258-6.html?sections=Comprehensive

In both of those, the 'population' is not defined. But in the LRI, it
was defined like this:
Reports the fraction of all of the reads at this genomic location that
were represented by the given allele. For homozygotes it will be close
to 1.0; for heterozygotes it will be close to 0.5. It can be a smaller
number when there are mosaics or multiple chromosome, or mixtures of
tumor cells and normal cells.

So, by the LRI definition, I think it makes sense as something to be
delivered with the variant observation? And I don't know that it
would need to be structured as its own Observation. I am not sure
that value "can reasonably be interpreted and used independently" as
stated in the Considerations for representing information using Observation.
Now, if we want to consider "Population Frequency", that does feel
more like "annotation / enrichment, interpretation, significance or
prediction" as Amnon said below. It seems to be a good and fairly
simple candidate for us to test the waters of delivering knowledge as
something else (a new resource, a profile on Observation, something
else?). As an aside, we need to allow the population to be specified as well.
For anyone curious about it, here is an example of where those
population frequencies might come from:
http://exac.broadinstitute.org/variant/22-46615880-T-C

--from Gil--
The STU3 definition came from Sequence Ontology (as a number of definitions did) based on group's selection in past doing it that the way where SO fit. I believe these were Bob M's use cases. Either way, they are a couple of CDS ones related to DAM. As part of recon process, LOINC codes were added where they may be closest fit last year (as in discussion here, it appears LOINC code was altered in the LRI later (but not on the site) and is used differently than the use in SO we selected). One solution is that LOINC code itself (or name of term) should be more clearly defined on the actual LOINC site. We can also revisit SO definition. For both SO and LOINC, there are also ways to make recommended changes (and we did it successfully in both cases in past).

--from Clem---
Think the definition was shortened by accident in the transfer to the FHIR model. -which change I had not noticed. So as Gil suggests we should get it to be its original intention. Clearly there are two different concepts that belong as separate terms. And if the population based one is needed it should be a separate a new attribute (Not sure what resource these "external knowledge" things are supposed to live. )

--from Kevin---
it feels like we need to do the following:
• Ensure we are comfortable with the wording from the LRI, or come up with common wording and update the LRI
• Ask LOINC to refine the definition
• Update FHIR definitions to match
• Optionally, ask SO to refine their definition (or, decide we are just breaking with their definition since it is a little too vague?)

Once we make that decision, we could then talk about adding "Population Frequency" - however I would suggest we hold off on that for now, as that really does get into the 'knowledge' realm, not 'observable', and I am not sure our group is really ready for that just yet.

--from Gil---
OK with me. I would just recommend that workgroup decide on what the
term name should be because many databases use 'allele freq' for what
some are thinking is pop allele freq (NCBI, Broad, SO,etc). They
specifically use and define "allele freq" that way. While we may be
able to get some to change, may be hard to change all. So, one option
is to think of a modifier like sample allele freq or something else..