Commentary on obographs proposal (@kltm)

kltm commented 7 years ago

@cmungall I'm unsure of the format that you want, so I'm making an actionable super/epic (bleh); I can split these up as well if that works for you.

[ ] The bbop-js codebase is /almost/ completely legacy/defunct, and completely defunct as far as graphs go. I've tried to port the wiki pages and links over; the proper base reference should be: https://github.com/berkeleybop/bbop-graph https://github.com/berkeleybop/bbop-graph/wiki http://berkeleybop.github.io/bbop-graph/doc/index.html (although the "doc" instead of "docs" is probably a typo I should fix) https://www.npmjs.com/package/bbop-graph
[ ] While the etymology is muddled, I'd avoid the "BOG" acronym ;)
[ ] I'd note that in your example you make use of the meta tag. The idea for bbop-graph here was to allow (in the future namespaced) subclass extensions. If you need to have that structured better, we can move our extensions elsewhere and you could mark that as reserved for other use in your spec.
[ ] I was wondering about the type statements (looking at https://github.com/geneontology/obographs/blob/master/schema/obographs-schema.json). In the extended (subclass) bbop-graph model that noctua uses (bbop-graph-noctua), type statements look like:
```
"type": [
         {
           "type": "class",
           "id": "CL:0000312",
           "label": "keratinocyte"
         },
```
[ ] As well, we allow for compound and nested types; you seem to just have a choice of three?
[ ] I was wondering about the caps enums: "enum" : [ "CLASS", "INDIVIDUAL", "PROPERTY" ]. We use lowercase.

[ ] We have annotations on entities like:

"annotations": [
      {
        "key": "contributor",
        "value": "http://orcid.org/0000-0001-7476-6306"
       },

[ ] I think ideally we could work out programmatic reusability, even if the exact specs were a little different. For example, we have quirks where we allow nested graphs and the like, which is not relevant for your use case.

cmungall commented 7 years ago

Thanks, some early thoughts, more later

github links

thanks, we should fix the links

basic profile

OK, we should spell out the BOG profile :-) though it's mainly there for explanatory purposes, there is no distinct format, it's a profile like OWL2-EL etc

meta tags

The goal here is to have a fixed standard that can express the full range of OWL, with some convenient shortcuts for common idioms.

So would it make sense to have the obograph meta be an official subclass in bbop-graph

types

This needs documented better. There is no nested type in obographs.

ClassAssertion maps to a graph edge using the pred id 'type', e.g:

https://github.com/geneontology/obographs/blob/master/src/test/resources/abox.owl#L57-L58 -> https://github.com/geneontology/obographs/blob/master/examples/abox.yaml#L18-L20

We could have opted for inlining the ClassAssertion into a property of the Individual node as in bbop-graph-noctua, but that would obscure the graph structure of the overall set of axioms (which is advantageous in noctua as it's an abox editor).

Now, we also use (inlined) 'type' as a key in nodes to indicate the type in the RDF full sense. Perhaps it would be better to call these something like 'declaration' consistent with OWL-DL. These are restricted to a set of datatypes/metaclasses.

Related #11 and #7

Annotations

Maybe we should align technology, key=pred, value=val. Pred was chosen for alignment to RDF technology and to emphasis that this is a declared predicate rather than an abritrary dict key.

Example:

https://github.com/geneontology/obographs/blob/master/examples/nucleus.yaml#L173-L178

Reusability

I'm hoping that the shared intersection of bbop-graph + obo-graph is sufficient to support basic graph viz. I expect for other cases a simple translation layer would be required.

kltm commented 7 years ago

It would be interesting to see practically how incompatible obographs and bbop-graph-noctua really are (i.e. are they both "subclasses" of bbop-graph or can they be related to eachother more directly).

Also, just to note that bbop-graph-noctua has more value-type as well:

                    {
                        "value-type": "IRI",
                        "value": "gomodel_taxon_559292-5525a0fc0000001-ECO-00000
05-553ff9ed0000001",
                        "key": "evidence"
                    },

kltm commented 7 years ago

Is the coverage of this ticket considered in https://github.com/geneontology/obographs/milestone/1?

geneontology / obographs