Add Biolink Model validation depth to Edge attribute (constraint) validation

[RichardBruskiewich]

The current TRAPI Knowledge Graph edge validation falls short of detailed validation of edge attributes (see attribute validation code snippet (or lack thereof) in the edge validator.

Deeper validation of Biolink Model (and perhaps, some TRAPI-specific) validation is required at this level.

[RichardBruskiewich]

Commentary from @edeutsch specifically regarding validation of the attribute_type_id (preferably inheriting from biolink:association_slot)

The constraint is not really absolute, mostly because many of the concepts we want to convey are not in biolink. So I think we should enforce a curie, so if there's not colon, then that should be an error. If the prefix is not biolink, that should be a warning. If the prefix is not known to biolink that should be a double warning. Are we able to report warnings? or just errors?

• ERROR if not a CURIE
• WARNING if not an instance of biolink:association_slot (but is another Biolink class?)
• (double) WARNING if prefix is not known to Biolink (i.e. not known to Biolink in what context? All registered prefixes?)
  _

[sierra-moxon]

toolkit.get_element_by_prefix() might help
if that returned an empty set, you'd know that biolink doesn't register the prefix on any of its classes.

If the double warning is instead to validate that the curie prefix is a valid prefix in the world, I think you can use other packages for that (see bioregistry).

For infores validation, I'd like to add BMT methods that would query that controlled vocabulary. We could repurpose this ticket for that purpose, if you think toolkit.get_element_by_prefix() will work for this.

[RichardBruskiewich]

Yes, toolkit.get_element_by_prefix() will suffice. I've added it with unit tests.

[RichardBruskiewich]

Substantially resolved by PR #31.

The one missing aspect of this issue to be implemented is the ERROR / WARNING / (double) WARNING classification above. This intent is covered by PR #27.

[RichardBruskiewich]

@sierra-moxon , @edeutsch , rather than opening up a fresh issue, I note our discussion today (23 March 2023) as discovering that toolkit.get_element_by_prefix() is limited to searches on category class id_prefix values and rather that the use case requires that we cast a wider net, perhaps including association slots. In the meantime, it almost feels like we need to revisit the Biolink model for biolink:Attribute (which has some relevant but not all id_prefix mappings), and its relationship to biolink:has_attribute_type, etc. as well as internal 'attribute types' like biolink:knowledge_source++ provenance.

[RichardBruskiewich]

    "warning.knowledge_graph.edge.attribute.type_id.unknown_prefix": {
      "CHEMBL.COMPOUND:CHEMBL112--biolink:occurs_together_in_literature_with->CHEMBL.COMPOUND:CHEMBL1201558": [
        {
          "attribute_type_id": "EDAM-DATA:2526"
        },
        {
          "attribute_type_id": "EDAM-OPERATION:0226"
        },

[RichardBruskiewich]

  attribute:
    is_a: named thing
    mixins:
      - ontology class
...
    slots:
      - name                   # 'attribute_name'
      - has attribute type     # 'attribute_type'
      # 'value', 'value_type', 'value_type_name'
      # extracted from either of the next two slots
      - has quantitative value
      - has qualitative value
...
    id_prefixes:
      - EDAM-DATA
      - EDAM-FORMAT
      - EDAM-OPERATION
      - EDAM-TOPIC

[RichardBruskiewich]

biolink:has_attribute_type is defined:

  has attribute type:
...
    domain: attribute
    range: ontology class
    multivalued: false
    required: true
...

[RichardBruskiewich]

If anything, all the TRAPI attribute_type_id mappings ought to be here, but aside from having no parent (what should it be? An (association or node) slot?, its range is biolink:OntologyClass which may feel sensible except for the fact that some of the acceptable attribute_type_field values are not really ontology terms (I mean, the biolink:association_slot child biolink:primary_knowledge_source is not a biolink:OntologyClass?) Maybe I'm being pedantic but then again, maybe we need to clean up the model here!

[sierra-moxon]

this test in bmt works:

    # get class by prefix
    elements = toolkit.get_element_by_prefix("EDAM-DATA:123345")
    assert "attribute" in elements

as does this:

    # get slot by prefix
    elements = toolkit.get_element_by_prefix("BSPO:123345")
    assert "original predicate" in elements

so I think the method is working on all elements at least

[sierra-moxon]

Also, technically, biolink is in BioPortal as an ontology :) - but I agree the attribute in biolink does not map to a TRAPI attribute.