[OBO-Edit Working Group] last transcript - oboedit, protege, reasoners

[chris m. <cj...@fr...>]

I just want to correct a few factual errors and insert a few personal  
opinions regarding the last transcript (was on holiday last week)

- it seems the transcript dived in with a detailed discussion of  
reasoners, possibly leaving a few people wondering exactly what the  
reasoner was and what it did. John now has some excellent  
documentation on how the reasoner works. we could still do with more  
docs on *why* you'd want to use the reasoner. this will become  
apparent with cross-products (btw, regulation may be a good place to  
start)

- protege doesn't "have a reasoner" per se - protege can use 3rd  
party reasoners like pellet, but this isn't part of protege as such.  
oboedit *could* use 3rd party reasoners, if John decided this would  
be a better option (this is point 7, which John says is difficult)

- I think John is talking down his reasoner. It lacks many features  
of standard OWL reasoners - however, these are features we don't  
need. John mentions the lack of reasoning over symmetric relations in  
the oboedit reasoner. But we don't need this yet! If GO or another  
OBO ontology requires this, we'll add a feature request, and it will  
be prioritised appropriately. It seems that the current reasoner  
fulfils the requirements of GO and other OBO ontologies for cross- 
products (but we have to fully test to be sure)

- John, the FMA has a MySQL database because it's fairly large (but  
not *that* large), not because they need it to do reasoning over the  
FMA. In fact the FMA maintainers use Protege-frames, not Protege-OWL,  
so there's not a lot of useful reasoning you can do on it. Now some  
people have converted the FMA to OWL and added logical definitions  
and demonstrated reasoning to check for inconsistencies, but this is  
a research project. At this point I may have left a few of you  
behind: what, there's *2* Proteges? What's the difference? How do I  
choose one and not the other? Can I use the same files? No? Well,  
yes, but there's issues? these are tricky issues and IMHO yet another  
reason oboedit is a superior tool for people who don't have time to  
get distracted in some gnarly CS issues.

OBOL:

- in actual fact, Obol works well on a lot of non-GO ontologies

You can test the obol web interface - see http://www.fruitfly.org/ 
~cjm/obol for details

I'm keen to have Obol integrated into oboedit, but I want to make  
sure what the requirements are before going ahead and prioritising this?

First of all, Obol is fairly slow. This is fine for getting  
suggestions for logical definitions for a single term

scenario:

curator creates a new term "microglial cell activation". They type in  
the term name. Obol is called and comes back with a logical definition:

   "cell activation" WHICH has_specific_outcome "microglial cell"

ie cell activation is the genus, the part after the WHICH is the  
differentium

This saves the curator the effort of using the xp editor to build the  
definition by searching CL etc (which may or may not be onerous,  
depending on how the xp editor works)

Then the oboedit reasoner tells the curator that micorglial cell  
activation" is_a "macrophage activation" - but this ONLY works if  
macrophage activation has a logical definition referring to the  
correct CL term (and CL is loaded of course).

It may make more sense to use Obol as a series of one-off runs,  
running overnight, generating candidate logical defs that are refined  
by curators.

Also, remember, even structured synonyms don't capture as much info  
as a logical definition - Obol will get things wrong, sometimes  
spectacularly so

if it will take John a long time to integrate Obol, the time may be  
better spent getting the xp editor really slick, making it easy and  
fast to edit the cross-products/logical definitions.

For more info:
http://berkeleybop.org/mediawiki/index.php/GO:Logical_Definitions

This page also has an example of another reasoner (Pellet) running in  
another editor (SWOOP, not Protege) which as its own explanation  
plugin (which is scary - I think oboedits will be better). This  
should hopefully put some context on some of the discussion I see at  
the beginning of the transcript.