[Obo-discuss] Making OWL versions of OBO ontologies available as linked data?

[Matthias S. <sa...@gm...>]

In the Semantic Web community, it is becoming increasingly popular to 
publish RDF/OWL following the linked data method [1]. Basically, this method 
requires that each URI of an ontology in OWL can be resolved via HTTP to 
yield a small RDF graph that describes the resources identified by that URI.
While the current URIs of OBO ontologies in OWL do resolve to RDF, the RDF 
returned consists of the whole ontology. For example, resolving the URI of 
an entity from the disease ontology such as
http://purl.org/obo/owl/DOID#DOID_0050043
resolves to a very large file containing the whole disease ontology. This is 
not practical for linked data clients, who need to be able to load data from 
ontologies in an incremental, entity-by-entity process.

Unfortunately, switching to linked data would also require changes to the 
URIs, because the part of the URI after the hash ("#"), cannot be used by 
the server. Furthermore, it might also require some thoughts about how 
OBO-in-OWL ontologies should be chunked. While the standard way of chunking 
ontologies for presentation as linked data is to return the RDF triples 
surrounding the resolved URI, some of the information in the OBO-in-OWL 
ontologies is _two_ triples away from the resolved URI (definitions, for 
example). A more sophisticated mechanism of chunking that includes these 
distant -yet important - triples might be of advantage.

I remember from personal communication that such ideas for OBO were already 
floating around for some time, but I have not seen any official announcments 
of such an undertaking. If the OBO community agrees that switching to URIs 
that resolve in a more selective fashion, than that should be communicated 
and implemented as soon as possible. The OBO ontologies are among the most 
important and stable open resources we have in the biomedical ontology / 
Semantic Web area at the moment, and they can act as a major backbone for 
inter-dataset mappings. A (perhaps necessary) change of the URIs will be 
quite disruptive now, but it will be even worse when uptake of OBO URIs has 
increased even further.

There are some interesting tools and projects appearing on the scene that 
make use of data offered following the linked data methods. The "Linking 
Open Drug Data" task [2] of the W3C HCLS interest group is one example. 
Having OBO ontologies available in this fashion would add a lot of value to 
these ontologies (from the perspective of these projects, at least).

[1] http://en.wikipedia.org/wiki/Linked_Data
[2] http://esw.w3.org/topic/HCLSIG/LODD

Cheers,
Matthias Samwald

DERI Galway, Ireland
http://deri.ie/

Konrad Lorenz Institute for Evolution & Cognition Research, Austria
http://kli.ac.at/