Meeting Minutes 20060601

Ontology Working group under NCBC-SDIWG

Daniel Rubin, Aris Floratos, Andrea Califano, Mark Musen, Zac Kohane, Ron Kikinis, Barry Smith, Karen Skinner, Ivo Dinov, Suzi Lewis, Peter Lyster (note taker)

6/1/06

ACTION:

1) a) “States 8 ontologies of significance” goes on Wiki; b.) every one mandated to comment and we will create a place on the wiki (Zac will send out announcement to SDIWG when web site is open for comments;

2) Things where people will hold their nose and use even though they aren’t perfect. Enumerate the next set of ontologies that may not be as well formed.

3) Where we need action which is currently unmet, e.g., ontology for phenotype and what will it take to develop that—we may pontificate proscriptively if we cannot afford to do, may want to find way to how to fund. Need technological vehicle.

4) Prepare for AHM. We have 2.0 hour meeting plus reporting so will be more endorse action and articulate publicly.

5) Suzi will put upcoming meetings on Wiki

MAIN MEETING MINUTES:

Zac: Summarize the current threads (not all on the phone have been aware of): what are the goals of this working group—do they pertain to NCBCs or does it pertain to wider audience? Second thread: what is the class of ontologies? Science ontologies and/or tools ontologies (the latter are covered by Yellow Pages/Resourceome, and so is out of scope). Re the former: David States gave us good strawman (someone should put it on Wiki?). One big issue arose out of this exercise: connection between GO and MeSH. Is NLM going in this direction? There seems precious few folk involved in this and it is of some interest to NCBCs. There was some issue of ‘formality of ontology’—the issue of merging (GO/MeSH) is non trivial. A goal of this tcon is to move forward to the All Hands Meeting (AHM). Are we going to adopt a set for NCBC-wide use, or adopt use of OBO Foundry style of approach (Suzi is interested), or engage in more formal semantic web effort? How to proceed so AHM is not just fact gathering?

Mark: Want to have minimal set, but how to take ‘affirmative action’ MeSH and ICE are oldest and thus difficult to extend (200 years span of decisions to go over). Is a dilemma. Short term goal: what to use short term. Long term: how to fix things.

David: UMLS is ongoing and we will have overlap with them. Perhaps we (NCBC) can focus on own needs and use UMLS as basis. Zac agrees. We don’t necessarily have important ‘votes’ but ‘eat our own dog food’—this is powerful sign of (1) usability and (2) utility.

Zac: Ask Mark how to map MeSH GO in principled way. Mark: UMLS? Zac wonders if semantic links are sufficient. UMLS has point-to-point but loses semantic links. Is there a process of MeSH+GO? Mark: What tools? Who pays? There may be a proposal for a wikipedia-type activity. Zac: Can we agree prior to AHM that this activity is good to help us annotate our data in genome-phenotype way.

Barry: MeSH was designed by librarians to annotate literature (i.e., just for indexing). Not necessarily used for high-grade scientific research, computer reasoning. Mark: Should not ‘bless’ MeSH to be anything other than medical subject headings. MeSH ‘ought’ to be useful for data integration and NLP. As NCBC we can include medical subject headings.

Is the need—define minimal terms for annotation, as opposed to ontologies. Agreed.

Ron: Interested in image classification (e.g., subvoxel classification). Where should we short term invest in commonalities?

David: maybe get most benefit by mapping subset of useful terms. Zac: not sure what is vehicle.

Barry: Mappings don’t seem to help (GIGO). Rather than mapping, think of enhancing. Mark: Overlap MeSH/GO is 20%. GO doesn’t describe pathological space. Everyone agrees that in the abstract it would be better (GO is small compared with MeSH it is difficult to have GO push MeSH).

Zac: for now, are we just ‘blessing’ terminologies? What additional efforts? We can’t make MeSH GO either go away or merge. Can we categorize ontologies into three categories.

(1) Ontologies that we endorse (reference ontologies);

(2) Ontolgoies we don’t ‘endorse’ but acquiesce to that cover our domains and are at forefront (MeSH ICD legacy ontologies),

(3) Area of uncertainty of unsure leadership—e.g., urgent need for phenotype ontology. Some desiderata in this effort.

David States: has done good Inventory—and really highlighted an important issue with MeSH. We should take action on David’s list.

Andrea: where does description of software come in? Zac that is important but perhaps off scope for this. Aris wants to know if we should treat as different problem. BISON is glue to help. Just keep this out of scope. BISON is more relevant to the Yellow Pages/Resourceome effort.

Action: 6) a) “States 8 ontologies of significance” goes on Wiki; b.) every one mandated to comment and we will create a place on the wiki (Zac will send out announcement to SDIWG when web site is open for comments;

7) Things where people will hold their nose and use even though they aren’t perfect. Enumerate the next set of ontologies that may not be as well formed.

8) Where we need action which is currently unmet, e.g., ontology for phenotype and what will it take to develop that—we may pontificate proscriptively if we cannot afford to do, may want to find way to how to fund. Need technological vehicle.

9) Prepare for AHM. We have 2.0 hour meeting plus reporting so will be more endorse action and articulate publicly.

10) Suzi will put upcoming meetings on Wiki

Peter: Need to distinguish between mode when you are ‘open’ and in ‘policy mode’ during the discussions.

David: AHM will be mostly presenting where we are at and where we can go.

Suzi/Mark: In terms of third bucket: upcoming meetings MOD phenotype, anatomy, clinical trials. Anatomy Seattle September, disease ontology November, open phenotype November

TTD: Put these minutes on the wiki? Everyone agrees.