Coordinate on strategy for atlas region URIs?

[cmungall]

With each uberon release we create bridge axiom ontologies for 5 allen atlases

• http://uberon.github.io/downloads.html#bridge

There are various importers that will pull in the uberon backbone and hang off the ABA classes

We have also started folding these into the composite metazoan build, see:
http://uberon.github.io/release/2016/06/12/release.html

A prequisite to this is a semantic representation of the atlases in obo/owl. Here is how it currently works, it's not sophisticated:

https://github.com/obophenotype/uberon/blob/master/Makefile#L1314-L1336

They are placed here:
https://github.com/obophenotype/uberon/tree/master/source-ontologies

Note that we translate to a partonomy. Years ago Allen published the MBA (then just "ABA") as an OWL file, but this was a subclass hierarchy, which only makes sense if you read each class X as somethig like 'regional-part-of-X'

We always use the structure_ids (as requested by the Allen people - https://twitter.com/tgilhou/status/531912044464066560). We prefix these with {D}{H,M,P}BA, and have these fall into the OBO base URI for now.

There are a number of issues here:

• the 5 prefixes are not officially OBO registered
• the ontology URIs should be resolvable (you need a catalog xml for the importer right now)
• the ontology class URIs should be resolvable
• the translation and publication should be separated from the uberon repo
• the Allen folks should be involved

It may be an idea to sync our requirements here first and approach the Allen folks. I'm guessing you will want URIs and probably a standard prefix for loading into your SG instance.

Note that previous neurolex minted entirely new IDs for brain atlas regions. While this is perhaps justified for systems that don't provide them, the IDs provided by Allen now seem stable, and they it would be better to use these in combination with a sensible URI prefix

One possibility is that we make these OBOs - we register each individually with the OBO Library (and would automatically be visible with the growing number of browsers that use this registry). Each atlas could have its own GH repo, or a shared repo, with a CI job or manual release manager making periodic releases, using the current uberon procedure, or a more robust variant of it.

cc @dosumis @mellybelly

[tgbugs]

Regarding the 5 issues.

1. Yes, we have been doing it mostly unofficially by making sure we match what uberon has.
2. I've been dumping our (currently in development) parcellations here.
3. I am currently using http://api.brain-map.org/api/v2/data/Structure/ as the underlying iri prefix for both HBA and MBA, which forces me to use ABA for the curie prefix. This is a problem that needs to be addressed in 5.
4. I have my general parcellation scheme generation here, I can adapt this to produce the desired output format if we can get prefix specific iris (ie HBA and MBA need to have different prefixes), or can just point to the files that are generated some other way (since you already have many of them). I also encounter this issue when dealing with the FSL atlases and if there is a common solution that would be very helpful and I agree that the neurolex solution is quite unsatisfactory here because there are indeed stable and globally unique identifiers, they just happen to resolve poorly and lack human readability. The obos solution seems reasonable, I am planning something similar for parcellations.py so a coordinated effort would be best.
5. Yes, we need to do this to address both 3 and 4.

[tgbugs]

Quick update on this. For the time being my solution is to use the following:

'HBA':'http://api.brain-map.org:80/api/v2/data/Structure/'
'MBA:'http://api.brain-map.org/api/v2/data/Structure/'

It is dirty and suboptimal, but it works with all the existing systems. One failing is that if a browser renders the url it will convert any hba identifiers in mba identifiers. The larger failing of course is that this only works for 2 identifiers. If they had a working https endpoint we could get 4, alas, not possible.

[cmungall]

Minor change - remove the :80?

[tgbugs]

The :80 is what prevents last-one-wins behavior from owlapi and friends, without it one of the prefixes will be converted to the other.

[cmungall]

oh right, OK, I couldn't quite understand what you were saying before, but obvious now, I see. Yeh, maybe worth pinging them to see if they can provide something, but if this works for you for now..