Proposal for enhancement to "id_prefixes"

[hsolbrig]

1) id_prefixes currently says "the identifier of this class or slot must begin with one of the URIs referenced by this prefix".

This sort of implies that a prefix can reference more than one URI. I'm hoping that we are dealing with a model where every prefix maps to exactly one URI (note, however, that the reverse may not necessarily be true... I need to check whether we guarantee uniqueness on URI's per prefix)

2) when it comes to actually validating data, I would think that the following:

classes:
    HighClass:
        id_prefixes:
            - NCIt
            - SCT

Would assert that a YAML or JSON representation of the id of an instance of HighClass would necessarily start with "NCIt:" or "SCT:", while an RDF instance would start with https://nci.....org/ncit/... or http://snomed.org/id/.

What I would propose, however, is that we extend the definition of id_prefixes to support the following:

classes:
   HighClass:
       id_prefixes:
          NCIt:
          SCT:

Which would be the same as the above. We would extend the definition slightly, to allow:

classes:
    HighClass:
       id_prefixes:
           NCIt: ^C\d{5,6}$
           SCT: ^\d{6,18}$

Which would assert that the local name of a Curie or URI must begin with "C" and have 5 or 6 digits if it began w/ NCIt or it must be a 6 to 18 digit number if it were SCT.

This would be a minimal change to the LinkML model itself, and, as of yet, the loaders do not do anything with ID prefixes so it would be no additions.

Questions: 1) Do we really need the "^...$" pattern or can we assume them? 2) Would we ever want two or more patterns and, if so, would we want something of the form

    id_prefixes:
        NCIt:
          - C\d{5}
          - M\d{7}

or would "(C\d{5}|M\d{7})" be ok?

3) It should be noted that SNOMED CT, in particular, includes a check digit and other formatting information that isn't expressible as a simple RE. Should we provide a hook for future use that names an algorithm or just let it slide.

My suggested answers are: 1) Assume them, 2) single RE is fine and 3) nah - not now

[cmungall]

See also https://github.com/linkml/linkml/issues/194

[cmungall]

1. I think this is a bug in the docs. It should just say "must begin with one of the URIs referenced by this prefix". Or we could frame it in terms of CURIEs.
   • let's avoid NCIT as an example though, as there are in fact different URI interpretations of the CURIE (OBO and native)
2. I like the general idea but some drawbacks:
   • people who are currently using id_prefixes will have to change their code
   • violates DRY. Some prefixes are 1:1 with classes. Others (NCIT, SCTID, KEGG) cover many classes, e.g in biolink. To complicate, some (e.g. KEGG have a datatype-specific localId pattern), others don't

[hsolbrig]

people who are currently using id_prefixes will have to change their code

Are you talking python code or yaml? The yaml is backwards compatible, but you are correct -- the python will change from a list a dictionary...

[hsolbrig]

violates DRY. Some prefixes are 1:1 with classes. Others (NCIT, SCTID, KEGG) cover many classes, e.g in biolink. To complicate, some (e.g. KEGG have a datatype-specific localId pattern), others don't

Hmmm... this makes me suspect that we may not be viewing this exactly the same way. What does it mean to "cover many classes"?

[cmungall]

See biolink

  pathway:
    is_a: biological process
    mixins:
      - ontology class
    exact_mappings:
      - PW:0000001
      - WIKIDATA:Q4915012
    narrow_mappings:
      - SIO:010526
      - GO:0007165
    id_prefixes:
      - GO
      - REACT
      - KEGG
      - SMPDB
      - MSigDB
      - PHARMGKB.PATHWAYS
      - WIKIPATHWAYS
      - FB  # FlyBase FBgg*
      - PANTHER.PATHWAY
...

  anatomical entity:
    is_a: organismal entity
    mixins:
      - thing with taxon
      - physical essence
    description: >-
      A subcellular location, cell type or gross anatomical part
    exact_mappings:
      - UBERON:0001062
      - WIKIDATA:Q4936952
      # UMLS Semantic Group "Anatomy"
      - UMLSSG:ANAT
    narrow_mappings:
      # Body System
      - UMLSSC:T022
      - UMLSST:bdsy
      # Body Location or Region
      - UMLSSC:T029
      - UMLSST:blor
      # Body Space or Junction
      - UMLSSC:T030
      - UMLSST:bsoj
      # Body Substance
      - UMLSSC:T031
      - UMLSST:bdsu
    id_prefixes:
      - UBERON
      - GO
      - CL
      - UMLS
      - MESH
      - NCIT
    in_subset:
      - model_organism_database

GO IDs are somewhat heteregeneous. They can be used for 3 bl classes (process, activity, anatomy/cell component)

NCIT IDs are very heterogeneous

[cmungall]

can we close this and move any outstanding issues to https://github.com/linkml/linkml/issues/194