Identifiers

[xrotwang]

The definition of Individual_ID seems inconsistent:

The Individual_ID column has to represent each sample with a world-wide unique identifier string equal to the identifier used in the respective accompanying publication.

If Individual_ID has to equal the identifier used in the publication, how can one make sure it's also a UUID? The specification goes on to say

There is no central authority to issue these identifiers, so it remains in the hand of the authors to avoid duplication. The Individual_IDs are also employed in the genetic data files and therefore have to adhere to certain constraints.

which might create even more problems to reconcile "identifier as in publication" with what Poseidon wants. From my experience, I'd recommend to distinguish local IDs and global UUIDs explicitly, and specify all of the constraints for the UUID.

[xrotwang]

Btw. the definition in the schema does not have the "world-wide unique" clause.

[stschiff]

OK, let me play dumb and naïve here. Why would it be so terrible if we simply enforce uniqueness within our tools once a set of packages gets read in? Why do we have to somehow enforce "world-wide-uniqueness" at the format specification level? So, in other words, why does "uniqueness of identifier" have to be something that we need to somehow guarantee globally, rather than locally?

[xrotwang]

I agree that "world-wide-unique" may be a bad trade-off. Often somewhat meaningful local IDs are a lot easier to maintain, and combined with some sort of prefix (possibly derived from provenance) when aggregating data, things work well. I'd still say that having identifiers for things that are referred to from other objects are an important piece of data - and should not only be minted upon reading data - and samples seem to be such a thing.

[stschiff]

Fair enough. Well, Clemens and I have pondered some kind of "Poseidon" sample ID, which we declare once a new package is taken into our central repository. That may make sense at some point.

[stschiff]

This was discussed again on Oct 14.

• AADR has the concept of Version ID vs Master ID, where the Master ID is arguably a unique identifier. We could adopt something like this, which we could then call Poseidon Master Individual ID or something more with a ring to it.
• Alternative would be to introduce a column called Related_To and Degree_of_relatedness which would act as key into the same table to describe relationships.

Preliminary decision:

• Change Individual ID to Poseidon ID
• Introduce AKA column which lists multiple other names the individual is known as (list column)
• Introduce Related_To column, also list
• Introduce Relation_Type column, also list, choice field something like Identical, 1st degree...