[MOBY-l] further thoughts on atomic services & service description

[Andrew D. Farmer]

Hi Mark-

this seems like an interesting direction to explore. My initial take is that
you'll always run into cases in which two natural datatypes can have two
or more natural associations, e.g.:
	sequences related as orthologs or sequences related as paralogs;
	references authored by a person or references about a person;
	genes that encode proteins or genes regulated by proteins;
	genes associated with GO terms based on experimental evidence vs
		genes associated with GO terms based on motif analysis
	etc., etc.

So at some level, it seems natural that there will be "meaning" attributed
in some way to the associations that can't be captured by just specifying the
datatypes on either end of the transformation. (I think the last example
of GO term-gene associations subclassed by evidence also suggests that one
could go crazy trying to define ontologies of relationship types...)

That said, however, I do agree that if one is providing a service
that is to be *understood* as conceptually equivalent to a path through a more
simply defined set of edges, then it might be useful for people to have the
ability to describe it in these terms, without necessarily mandating that
no one can define a relationship between datatype X and datatype Y unless
they describe it in terms of a canonical set of edges that will lead from
X to Y.

On the other hand, I'm not sure I see how this would all work out in practice.
I suspect that providers of services would tend to describe them as compositions
of more primitive services, only if that happened to be the way it mapped to
their implementation, but if not, they wouldn't tend to worry too much about
how their basic operation might be conceptually composed of primitives
someone else had defined? And I think we also want to be careful here to
distinguish between what is needed for conceptual clarity and what is really
"implementation detail"...

But I'm resolved to think about it some more, and I definitely think you should
keep pushing your thoughts in that direction.


PS.
It seems to have some interesting analogies to the premise of the ISYS data
model approach, where the "meaning" is to be found at the granularity of
"attributes", and objects can be more or less freely composed out of this
vocabulary... I suspect the analogy is fairly superficial, but I'll
think about it some more...



Andrew Farmer
adf at ncgr.org
(505) 995-4464
Database Administrator/Software Developer
National Center for Genome Resources