Skip to content

Latest commit

 

History

History
86 lines (55 loc) · 4.61 KB

README.md

File metadata and controls

86 lines (55 loc) · 4.61 KB
Raw

STATUS: EXPERIMENTAL

Aims

Generate a LinkML version of the Cell Annotation Schema that can be used to:

  • enhance validation with dynamic enumns (e.g. for CL) and internal reference checks (e.g. for CL).

    • longer term aim - build a better infrastructure for Link-ML dynamic enums using UberGraph (complete graph of OBO ontologies) & OBASK autosuggest endpoints. see linkml/linkml#1775 (comment)

  • generate OWL/RDF from data to drive knowldge graphs (the RDF is also potentially useful in cas-tools - e.g. for calculating cell_id closure.)

  • generate python data classes e.g. for use in cas-tools

The current approach is to incrementally build and test desired functionaly with a scratch version of the schema. Once major issues are solved we can move to a more complete representation and start using generators for doc, JSON-schema, python Dataclasses etc. At this point we should also automate tests against CAS compliant.

Because the need for an OWL/RDF representation is most critical for BICAN, initial work is concentrating on modelling the BICAN extended version of cas (snapshot release).

RDF schema sketch (based on prior work in pandasaurus_cxg, VFB, BDSO/PCL)

  • Annotation objects are instances of 'cluster' (PCL:0010001) # could be improved.
  • Clusters that form a simple nested hierarchy are related by subcluster_of (RO:0015003)
  • Clusters are typed with a Cell Type term using an existential resriction on composed primarily of (RO:0002473)
  • cell_label --> rdfs:label proposal: annotation.cell_fullname --> skos:preflabel?

TBD: cluster --> labelset - should this use internal CAS IRIs?

Files

Inputs:

  • schema/scratch/scratch.yaml # Exptl schema
  • schema/scratch/data.yaml # test data file

Outputs:

  • schema/scratch/data.ttl # ttl file generated from data
  • schema/scratch/scratch.json # JSON schema output

To run

generate json schema: gen-json-schema scratch.yaml > scratch.json

validate: linkml-validate -s scratch.yaml data.yaml

convert to rdf (ttl): linkml-convert -s scratch.yaml data.yaml -o data.ttl

Expand dynamic enums (via oaklib): vskit expand -s schema/scratch/scratch.yaml -o scratch_expanded.yaml CellTypeEnum

Notes:

The schema in scratch.yaml replicates the basic functionality needed to reproduce cas json schema.

It also goes some way to producing RDF following the standard annotation schema developed elsewhere (pandasaurus_cxg, VFB, BDSO/PCL):

  • annotations are represented as individuals typed as clusters and are related in a hierarchy via subClusterOf
  • the correct predicate is specified for linking to Cell Type (CL term)
  • Dynamic enum & CL term linking works as a simple triple (requires expansion first or rdf gen fails)

Still needs work:

  • CL term is not treated as an OWL entity (value type of axiom is string) - Relevant ticket: linkml/linkml#1775
  • Dynamic enums specifying a CL term in the CL slot. RDF conversion fails when this is in place, although oddly not validation. Relavant ticket linkml/linkml#1775 ?
  • Individuals are not being created for labelsets (no idea why when the same pattern is being used as for annotations)
  • There is no OWL typing of object properties (should be able to generate these from imports, so outside of LinkML)
  • Strictly Cluster --> CL term should be an existential restriction.
    • Is there some way to fix this with LinkML OWL?
    • Could be hacked post-generation with SPARQL update.

To explore

To complete

  • The whole schema could potentially be pulled across by using schema-automator but so far, failing to install. In the absence of this solution, it could be added manually.

  • Some simple OWL infrastructure & ROBOT pipelines could pull minimal imports (APs, declarations)