Re: [Obo-discuss] obo foundry - unique term labels

[Chris M. <CJM...@lb...>]

On Dec 22, 2010, at 8:16 PM, Matt wrote:

> Inline response here.
> 
> On Wed, Dec 22, 2010 at 7:49 PM, Chris Mungall <CJM...@lb...> wrote:
>> 
>> Hi Matt
>> 
>> The problem with application-specific labels is that they are application-specific. There is a requirement for a reliable, consistent generic application-neutral means of rendering a label for a class that won't cause confusion for a user of a system that incorporates multiple AOs. You *could* build clever display logic into every single application, but this wouldn't be very efficient.
> 
> By stating that "it's an application specific issue" I meant simply
> that application must choose how to reference/utilize an identifier.
> Since this necessarily has to happen at some level whether you choose
> labels as unique or identifiers as unique I don't think it's the issue
> that you make it out to be, or, more likely, I don't get it :).
> 
>> Technically it would not be difficult to centralize this logic. You could have a tool that imported n ontologies, and automatically generated alternate guaranteed-unique labels as required. This could be a central service provided by the OBO library.
> 
> Sure, mechanisms for generating new unique labels is easy, but I guess
> I just don't see the point? One of our rules of thumb in the HAO is
> "if at all possible don't introduce new terms that haven't been seen
> in the wild". New labels in the absence of context will confuse the
> hell out of our domain community.  I do see the point of a simplified
> namespace "resolver", something that says "hey, you're using HAO, that
> means your referencing something to do with bees wasps and ants, does
> that help you understand what 'process' means in this context?".
> 
>> I think what you are railing against is the notion that you as the HAO maintainer would have to stick "hymenoptera" in front of all your class labels, greatly pissing off your users.
> 
> No, this is in fact not this case at all. I'm all for a plurality of
> labels. I'm completely against a plurality of unique identifiers.  I
> could care less what labels are provided for a given concept, IMO they
> don't matter when the class is referenced.  The simplest "application"
> implementation is using a hyperlink.  <a
> href="http:/some_obo_purl/hao_12345">any label you want who
> cares?</a>.  The obvious use case being different languages linked in
> this manner.  A nice side effect from this very simple use case is we
> can go back and harvest *new* terms/labels for existing concepts.

I think everyone on this list would agree about identifiers and hyperlinks, but we're discussing labels. Harvesting new valid alternate labels/synonyms is great too - this makes users and text miners happy, I'm all for this.

Most users also like having a meaningful unique label for each ontology class. I have seen a few ontologies that don't enforce this, and it inevitably leads to problems (the HAO doesn't fall into this category - I think you care more about labels than you're letting on here).

Things get more difficult when the uniqueness constraint is moved up from an individual ontology to an integrated collection of ontologies. Fortunately there is a simple solution, which is to automatically prefix names that would otherwise lead to clashes.

> cheers,
> Matt
> 
> ------------------------------------------------------------------------------
> Learn how Oracle Real Application Clusters (RAC) One Node allows customers
> to consolidate database storage, standardize their database environment, and, 
> should the need arise, upgrade to a full multi-node Oracle RAC database 
> without downtime or disruption
> http://p.sf.net/sfu/oracle-sfdevnl
> _______________________________________________
> Obo-discuss mailing list
> Obo...@li...
> https://lists.sourceforge.net/lists/listinfo/obo-discuss