Aim 2.4A: Assessment of validation needs

[schristley]

Integrated data/metadata in the AK are complete and accurate and that integrated datasets returned from the query API are complete, accurate, and compliant with our adopted standards. An example of cross-data-type validation may require that AIR in mouse repertoires be assigned an epitope specificity assayed in mouse.

• Single class/field: required fields, data types, simple constraints between fields, controlled vocabulary.
• External references: ontology id resolution, pubmed id, PDB, etc.
• Multiple objects (cross-field, cross-datatype): relations between objects, i.e. a field that “joins” them like repertoire_id or v_call.

The above are validation rules for data to insure consistency but doesn't require biological knowledge.

• Biological knowledge as validation rules.
• If experimental protocol indicates locus (TRB) then all rearrangements v_calls should be from that locus.
• A TCR-pMHC complex requires TCRA/TCRB chains, a peptide, and an MHC. The TCRA, TCRB and MHC should be in the subject’s genotype.

Range of needs:

• We need to perform the same validation in multiple places, e.g. the GUI (data entry) and the (JSON, OpenAPI) API (import). Also consider ad-hoc python/R scripting.
• Might want to do certain forms of validation in our analysis pipelines, in particular on output files.
• Can we specify validation rules in a centralized, language-agnostic way then utilize them everywhere?

[schristley]

• Single class/field: required fields, data types, simple constraints between fields, controlled vocabulary.

The majority of these should be handled by schema validation. The AKC architecture will require that component repositories perform this validation before sending data. Enforced through OpenAPI3 specification of request/response data.