docs: add ontology resources and guidance for subject matter annotation

[bendichter]

Add helpful links and usage guidance for the Subject Matter section:

• Reference EMBL-EBI Ontology Lookup Service for searching ontology terms
• Clarify when to use Anatomy type (brain areas, cell types)
• Clarify when to use Disorder type (diseases, disorders)
• Reference Cognitive Atlas for cognitive functions using Generic Type

[bendichter]

I'm going out on a limb a bit here. I don't know ontologies that well.

[yarikoptic]

all of those should IMHO be

• linked at the level of dandi-schema model as/where appropriate
• Web UI of meditor should get a facility to construct a list of preferred available values out of already present in metadata records in the archive while allowing for entering new ones which would potentially validate as belonging to ontology tc

FWIW, somewhat of a similar approach, although IMHO a bit more structured since it is an entire consistent collection of records of various types (Institution, Person, ...), not just collection of dandiset records, is taken by the https://github.com/psychoinformatics-de/shacl-vue by our colleagues which @candleindark . E.g. go to https://dfg-cp-survey.trr379.de/ui/?sh%3ANodeShape=trr379cps%3AContributor&edit=true and try to enter affiliation -- it will give you a list of entries prepopulated from RORs index IIRC, but you would have a way to add a new entry for an institution:

[bendichter]

@yarikoptic this seems like an issue for DANDI Archive, concerning UX of the meditor, right?

[kabilar]

I'll defer to Yarik on this one. I am not yet too familiar with ontologies.

[satra]

yes this is a meditor issue, but also a sourcing issue on what ontologies/taxonomies can be used or are of relevance (UBERON, HOMBA, ICD-10/11, etc.,.). i'm pinging @tekrajchhetri as this can be exactly what structsense can do to improve metadata for the dandiset, and we will be pointing to a broad collection of ontologies through the bioportal api and other sources. yes, this is both a meditor issue and also the previous notion that editing some components of metadata are extremely tedious and should just be done with AI tools.

the final note i'll leave here is that there are a bunch of details that should be at the asset level and simply pulled up to the dandiset level. the design of the current models was to provide an easier way for individuals to enrich metadata at the dandiset level. ideally these should be at the asset level as well.

[tekrajchhetri]

Thanks @satra for pinging me. I think it would be a nice use-case given that we've now integrated the bioportal in StructSense.

[bendichter]

I should mention- the reason I am working on guidances around ontologies is because it came up as we are feeding this doc directly into AI to do this ontology mapping for us. That's why I am considering this a separate issue from the UX. I am mostly asking here if we can come to some agreement about the ontologies we recommend for different types of metadata.