Re: [Myexperiment-discuss] i need some workflows :-)

[Antoon Goderis]

On Thu, 5 Apr 2007, David De Roure wrote:

> BTW These are the ones I'm using now, which I found with google.  I guess
> what I'm looking for is a reference set we can all use for test purposes.
>
> http://www.chembiogrid.org/projects/proj_workflows.html
> http://biomoby.open-bio.org/CVS_CONTENT/moby-live/Java/docs/taverna/guide/workflows.html
> http://bioinfo.mpiz-koeln.mpg.de/araws/workflows
> http://mips.gsf.de/projects/plants/PlaNetPortal/taverna_workflows.html
>
> This has also got me thinking about what workflows should look like to be
> indexed usefully by google - and what the IDs should look like.  For

OK, well here are my immediate thoughts

I don't think Google can sort all our discovery needs, but I agree it's
what makes people's workflows visible.

Of course, to serve Google's crawler and PageRank metric, workflows should
look like hyperlinked Web pages. The alternative is to register
the workflow repository as a library of programs with Google Code Search.
However, one can't add custom programming languages yet, i.e. Scufl, so
you'd have to start thinking of publishing these as Java or Perl; let's
not.

So let's have workflows as auto-generated pages with a mixture of incoming
and outgoing links (a social network of services, ahum). Whatever's
published on myExp can be linked symmetrically but there are
unidirectional things like outside pubs or wet lab references. A quick
list:

symmetric
--------
authored wfs [person/group web pages]
best of lists [popularity/uptime]
pages describing constituting services
provenance logs

outgoing
--------
related pubs
outside data set ref
disorder under investigation
funding agency
author web site

As for the ID question, would you be using the same scheme for workflows
and data items? Different purposes require different schemes so I'd guess
we need a way to have multiple ones and navigate between them, i.e.
combine

- quotability (eg short DOI vs long LSID)
- memorability (eg author/disease based names vs numbers)
- cost of issuing (eg the ProteinDB struck a deal for issuing DOIs for PDB
entries), also in light of versioning/workflows expiring
- local vs global uniqueness/reconciliation. Kepler have run into this one
big time with issuing LSIDs at almost all levels. If you are looking at
archives/naming schemes, have a look at their KAR system
(http://kepler-project.org/Wiki.jsp?page=KeplerObjectManager).

Good luck :-)

Antoon