HCLS/OntologyTaskForce/Telcons/2006/03/07

Teleconference Minutes

Date

07-March-2006

Participants

John Madden (Duke)
Vipul Kashyap (Harvard/Partners), part of call
Joanne Luciano (Harvard)
Alan Ruttenberg (Harvard)
Xiaoshu Wang (MUSC)
Daniel Rubin (Stanford)
Nigam Shah (Stanford)
John Barkley (NIST)
Brian Gilman, on IRC for part of call

Planned Agenda Items

Completely devote discussion to use case identification.

Discussion minutes

We discussed the choice of primary use case for demonstration of utility of ontologies in HCLS.
BioRDF group is very interested in use of RDF markup to support aggregation/navigation. But we need to show that over and above navigational links, there is added value in applying "reusable" semantic links (which is to say ontologies, knowledgebases).
Alan Ruttenberg outlined a use case that bridges bioRDF interests and OWG interests, and also has "bench-to-bedside" potential.
- Whole Brain Atlas (WBA) is a site that contains lots of anatomic data and images about neurologic diseases.
- OMIM contains lots of knowledge about genetics of neurologic diseases.
- But WBA doesn't know any genetics, and OMIM doesn't know any anatomy.
- So wouldn't it be cool to formalize some "bridging" knowledge that connects genetics and anatomy.
  - A really simple bridge between the two might just be some simple knowledge that there is a disease that is mentioned on both sites, albeit possibly under different names.
  - More complicated bridging knowledge is conceivable. For example, we could represent the knowledge that two diseases, each mentioned on both sites, exhibit simiilar clinical symptoms.
Once some bridging knowledge was formalized (i.e. made into an ontology), we could use it facilitate generating "interesting" links between the two sites--links that reflect knowledge that is not "natively" contained on either site.
- The first step in this project would be to choose some knowledge for the ontology. Alan suggested, and all agreed, that the knowledgebase should be small, something like 50 triples. In principle, it could be plain RDF, RDFS, or OWL; but since we are looking to simulate some persistent, reusable, and extensible semantics, it would probably make more sense to make it RDFS or OWL.
- Once we have the formalized knowledgebase, we could implement the links between

   WBA <-> OurKnowledgebase <-> OMIM
 in a number of ways.

The least interesting would be to take our knowledgebase with us and approach the authors of the OMIM site and of the WBA site, and ask them to put links to entitites in our knowledgebase-namespace on their pages. This would be akin to:

   WBA --> OurKnowledgebase <-- OMIM

But that would be a pretty bland approach. Better would be if we set up our own site, and WE take the responsibility to create the links ourselves, without asking anything of the authors of WBA and OMIM, viz.:

   WBA <-- OurKnowledgebase --> OMIM

This raises an interface-level implementation question, namely:
- *IF** we put triples up on the web somewhere that ADD knowledge to what already exists on other people's site(s)
- *THEN** what's the most appropriate way to let users to access that additional knowledge transparently?
You could conceive of a lot of possibilities.
- Our website looks like a frame around the two websites we are linking.
- Our website is a popup that floats over the websites we are linking.
- Our website "hijacks" links that already exist on one or both sites.
- etc.
  - We need to think long and hard about this !!

Assigned Tasks

||Task||Responsible||Due||
||Task||Responsible||Due||

IRC Transcript

** You have joined the channel
** gilmanb (gilmanb@128.30.52.30) has joined the channel
** xwang4 (80172404@128.30.52.23) has joined the channel

JohnMadden hi xiaoshu
xwang4 hi john
xwang4 john, did u get my eamil about the writing idea?
JohnMadden yes I did and I have started reading it, but haven't finished
JohnMadden I have quite a few comments
JohnMadden I was thinking of doing them as pdf comments
xwang4 good, would love to read it
gilmanb hello gents
gilmanb quick question for you guys
gilmanb do you use tools in house that convert rdbm's -> rdf?
xwang4 me? I didn't do that
gilmanb ok
gilmanb thanks
xwang4 is there a standard tool for that ?
gilmanb there are a few tools out there but nothing standard
xwang4 should be SQL to RDF, can be standardized, I think
gilmanb yes that exists
gilmanb ddl -> rdf does not
xwang4 b/c it takes arbitrary transformation?
JohnMadden Hi Brian
JohnMadden are you on the phone too?
gilmanb sorry no I'm not
JohnMadden on the phone we have John Barkley (NIST), Vipul, Xiaoshu, Brian, AlanR, NigamShah (SMI), JoanneL
gilmanb I just got back from vacation
JohnMadden we're talking about use cases
gilmanb trying to catch up
gilmanb ok
JohnMadden currently talking about alanR's OMIM use case
gilmanb let me hop on
JohnMadden http://www.med.harvard.edu/AANLIB/home.html
JohnMadden this is the link Alan is talking about
JohnMadden how would we markup a page like whole brain atlas
JohnMadden the issue is that we need term url's to link to
JohnMadden we are talking about umls
JohnMadden the notion would be to link to umls terms represented as url's
JohnMadden the problem is finding a good vocabulary
JohnMadden Dan Rubin joined call
JohnMadden Vipul left
JohnMadden so alan's idea is to liknk together wholeBrainAtlas to OMIM using rdf links to a controlled vocabulary represented as a uri
xwang4 yes, it is a good usecase
JohnMadden could we use grddl?
xwang4 yes, I think this brain atlas ->GRDDL ->RDF->OMIM
JohnMadden http://smi.stanford.edu/projects/cbio/mwiki-internal/images/7/75/Gupta_disease_ontology.pdf
JohnMadden so one practical use case is semantic navigation
JohnMadden another is data exploration
JohnMadden One idea is "parkinson's has Lewy bodies, show me other diseases that have Lewy bodies"
JohnMadden again the question is whether the originator of the data must host the markup
JohnMadden or whether 3rd parties should create the markup and reference the website
JohnMadden nigam says we must go beyong searching
JohnMadden xiaoshu is talking about aggregation
JohnMadden versus the questioon of inferencing
JohnMadden you can consider the we itself as a knowledge base against which we can inference
JohnMadden (xiaoshu)
JohnMadden alan: I can use SPARQL already to do queries

** em (em@128.30.52.30) has joined the channel

JohnMadden right now that queries pellet
JohnMadden it does not query across sites
JohnMadden should we look at a repository model, then, or should we aim to look at a cross-web query
JohnMadden is biodash a good example of the kind of aggregation we are talking about?
JohnMadden http://www.w3.org/2005/04/swls/BioDash/Demo/
JohnMadden bridge ontology is an excellent use case
JohnMadden for this group
JohnMadden show a very small bridge ontology that bridges the brain diseases on omim and wba
JohnMadden this could consist of say 50 links
JohnMadden we could focus this more closely by giving a clinical context
JohnMadden or a user context
JohnMadden to select what is relevant and what is not relevant to our brdige ontology
JohnMadden in this way we could keep the bridge small
JohnMadden "find me diseases in which old men had weakness in their lower extremitites"
JohnMadden portal application-- what would it look like
JohnMadden great call, everyone
JohnMadden thankx

HCLS Index Page >

OTF Homepage >

Discussion Archive >

Post to HCLS listserv >