HCLSIG BioRDF Subgroup/Tasks/URI Best Practices/Issues
Issues Related to Use of URIs
- Should a URI point to a real world entity? Not always possible, as not possible to point to my genes. It also may not be practical to point to every possible instance
- Should a URI point to a document about the entity of interest?
- Should a URI point to the concept of the entity, e.g. the concept of a particular gene.
- If there is a URI for a gene what does it represent, e.g. the gene sequence, the structure, an image?
- What should a URI return? What is the contract?
- Should a URI always point to RDF?
- What is the best way to support versioning?
- Matthias Samwald: "If there is a URI for a gene what does it represent, e.g. the gene sequence, the structure, an image?". Bold answer: A URI for a gene represents a gene, a URI for a gene sequence represents a gene sequence, a URI for an image represents an image. All of these things are different, why not just make different classes for each of them? What we should really avoid is inconsistency -- like having a class for a 'gene' which is actually just an 'document of interest about a gene'.
Of course, we can have a philosophical discussion about what a gene actually is, but this does not hinder our modelling efforts in OWL to the least. If we have different notions about a gene (e.g. a piece of DNA with some functional properties, any inheritable unit, a whole set of processes that happen around the DNA etc.), we can make different classes for all of these different notions. In many discussion surrounding the use of OWL people make the argument that we often cannot decide on a clear meaning for a concept, as there are many different opposing views. These people don't see that this does not hinder the use of ontologies -- it is actually the reason why we are using them! We can have multiple classes (e.g. gene-1, gene-2, gene-3) to disambiguate what we are meaning -- something that is not possible with a purely linguistic approach.
- Matthias Samwald: "Should a URI point to a real world entity? ... It also may not be practical to point to every possible instance. Should a URI point to the concept of the entity?" I think I have found a solution to these questions in the ontology I am developing (link). These question are really fundamental for ontology development in the life sciences, but their significance is still underestimated.
Very briefly, my solution is the following: 1) There are two fundamentally different top-level classes in the same ontology: 'spatio-temporal-particulars' (particular things located in a particular space and time) and 'concepts' (particular conceptualisations without a given location in space and time). The spatio-temporal-particular can be annotated with concepts. 2) In many cases it is not practical to point to every possible instance. This problem is adressed by using 'exemplary' particulars. Example: '<liver-tissue> <has-exemplary-part> <some-liver-cell>'