Replace types used in place of identifiers

[mslw]

Current state

As of now, the Dataset class uses the following IRIs to define its slots (slot name formatting came from linkml and seems related to how the slots were defined - through attributes or slot usage):

Code snippet

``` python from linkml.utils.schemaloader import SchemaLoader schema = SchemaLoader("src/sfb1451_schema.yaml").resolve() for s_name in schema.classes['Dataset'].slots: slot = schema.slots[s_name] print(f"| {slot.name} | {slot.slot_uri} |") ```

slot	uri
Dataset_name	schema:name
Dataset_title	schema:title
Dataset_description	schema:description
Dataset_doi	bibo:doi
dataset__crc_project	schema:ResearchProject
dataset__version	schema:version
dataset__sample[organism]	openminds:Species
dataset__sample[organism_part]	openminds:UBERONParcellation
dataset__keywords	schema:keywords
dataset__license	schema:license
dataset__homepage	schema:mainEntityOfPage
dataset__last_updated	schema:dateModified
dataset__data_controller	dpv:hasDataController
dataset__author	schema:author
dataset__funding	schema:funding
dataset__publication	schema:citation
dataset__hasPart	dcterms:hasPart
dataset__used_for	prov:hadUsage

Three of them are types (or names of controlled terms), rather than attributes, in their respective vocabularies. I think that's bad (a property should be a property, and the type should be its range). These are:

• crc-project: schema:ResearchProject
• sample[organism]: openminds:Species
• sample[organism-part]: openminds:UBERONParcellation

How to annotate these properly?

I think one example we have is a data controller. We use the dpv:hasDataController property, and define a DataController class (that uses the dpv:DataController URI, but has a set of mandatory/optional fields defined by us).

Option 1: with linkML we can define any Dataset property we want (e.g. contributingProject, studiedSpecies, studiedOrganismPart or ) and define corresponding classes as their expected Range. The properties would then use our identifiers, and the classes would (likely) use those defined above (schema:ResearchProject, openminds:Species, openminds:UBERONParcellation), or at least refer to them as closely related.

Option 2: use a very generic property a'la "related to" from some standard vocabulary, and only use types to distinguish. So a dataset would be related to a given project, and a given species, etc. I think this is semantically valid, but ugly for processing.

Option 3: find the right properties in a known dictionary. I think it can be hard - at least I failed.

[mslw]

A snapshot from a whiteboard discussion, to refresh our memories:

Not included on the board:

• a generic catalog wouldn't know what to do with sfb1451:studiedSpecies, but it might know what to do with prov:derivedFrom
• many of dataset properties can be described in terms of provenance (even "0000955" being generated from "brain" through a manual lookup performed by myself, should we want to be that detailed...)
• we don't need to be very precise in terms of definitions
• we can be very precise, and use sfb1451 specific definitions (sfb1451:studiedSpecies) but then it's likely that the data can only be understood by us (within the sfb145 context)