Skip to content

Latest commit

 

History

History
executable file
·
108 lines (68 loc) · 5.67 KB

README.md

File metadata and controls

executable file
·
108 lines (68 loc) · 5.67 KB

Join the chat at https://gitter.im/INCF/neuroshapes Build Status GitHub release

Note

This is the new simplified structure of Neuroshapes. A v0 branch is available to work with the previous schemas format.

Welcome to Neuroshapes

The goal of Neuroshapes is the development of open, use case driven and shared validatable data models (schemas, vocabularies) to enable the FAIR principles (Findable, Accessible, Interoperable and Reusable) for basic, computational and clinical neuroscience (meta)data. The data models developed thus far entities for electrophysiology, neuron morphology, brain atlases, in vitro electrophysiology and computational modeling. Future developments could include brain imaging, transcriptomic and clinical form data, as determined by community interests.

Table of contents:

Goal

The main goal is to promote:

  • the use of standard semantic markups and linked data principles as ways to structure metadata and related data: the W3C RDF format is leveraged, specifically its developer friendly JSON-LD serialization. The adoption of linked data principles and JSON-LD will ease federated access and discoverability of distributed neuroscience (meta)data over the web.

  • the use of the W3C SHACL (Shape Constraint Language) recommendation as a rich metadata schema language which is formal and expressive; interoperable; machine interpretable; and domain agnostic. With SHACL, (meta)data quality can be enforced based on schemas and vocabularies (easily discoverable and searchable) rather than being fully encoded in procedural codes. SHACL also provides key interoperability capabities to ensure the evolution of standard data models and data longevity. It allows to incrementally build standard data models in term of semantics and sophistication.

  • the reuse of existing schemas and semantic markups (like schema.org) and existing ontologies and controlled vocabularies (including NIFSTD - NIF Standard Ontologies)

  • the use of W3C PROV-O recommendation as a format to record (meta)data provenance: a SHACL version of the W3C PROV-O is created.

Also, Neuroshapes aims at creating a community for an open and use case driven development of not only data models (schemas and vocabularies) and tools around them but also guidelines for FAIR neuroscience (meta)data.

Tutorials

Please check out our tutorials:

Adoption

The following projects have adopted Neuroshapes:

Formats and standards

All schemas in this repository conform to the W3C SHACL recommendation and are serialized using JSON-LD. For practical reasons, the defined schemas are combined in an envelop (an ontology actually) that conforms to Nexus KG schema format.

Testing shapes

These tests are going to validate that schemas conforms with the SHACL specifications. And if you add examples you want the tests to validate you need to provide them inside a directory called valid or invalid as depicted next:

|-- shapes
|   |-- neurosciencegraph
|   |   |-- datashapes
|   |   `-- commons
|   |       `-- list
|   |           |-- schema.json
|   |           `-- examples
|   |               |-- datashapes.json 
|   |               `-- valid
|   |               |   `-- recipe_ingredients_list.json 
|   |               `-- invalid
|   |                   `-- recipe_missing_ingredients.json
|   `-- prov     
`-- ...

Tests are proved to run test with python > 3.6. To run them follow next:

# create your virtual environment and install these requirements:
pip install pytest pyshacl
# fix owlrl isusue by removing the .py extension 
mv {virtual-env-path}/bin/owlrl.py {virtual-env-path}/bin/owlrl
pytest tests

More about the issue with current released version of OWL-RL on PyPI.

To test an specific shape, add a parameter --scan_dir where the shape is, for example:

pytest --scan_dir=../neuroshapes/shapes/neurosciencegraph/commons/list/

Roadmap

  • Creation of an INCF/neuroshapes Special Interest Group
  • INCF endorsement as a standard and best practice that support FAIR neuroscience data
  • Extension of the current data model specifications

License

The license for all schemas and data is CC-BY-4.0.