LinkedReproducibility

Hashtag: #LinkedReproducibility

Twitter: https://twitter.com/hashtag/LinkedReproducibility

• Data Science
• Knowledge Engineering > Linked Data
  • Units > RDF and Units [QUDT,]
• Information Systems
• ELI5: Our data is probably already aware of a cure.
  • Cure: Win/Win solution

Note

This page (linkedreproducibility.rst) is licensed with CC0 1.0 (Public Domain).

Please do implement these ideas and specifications.

– @westurner

StudyGraph: Document Nodes and Link Edges

We should use annotations with typed, reified edges to link between various studies with comparable and incomparable analyses. (e.g. OpenAnnotation OA RDF OWL with more data than threaded comments).

• PDFs are the de-facto standard for scientific publishing
• Many journals also / instead request HTML
  • PLoS, for example
• Then, a “study” is a document
  • title – schema.org/name
  • authors – lead first (usu.)
    • schema.org/TODO – author, creator, contributor
  • abstract
  • tags/labels/keywords are edges to a tag/label/keyword node
    • hierarchical
      • MESH
      • PyPI Trove Classifiers
    • folksonomy
      • tags
        • tags often require deduplication / part of speech normalization / de-pluralization / etc
    • Concept URIs
      • wikipedia, dbpedia, wikidata, etc
  • links to Linked Data
    • https://schema.org/Dataset
    • CSVW
    • StructuredPremises: Premises as structured data
• What we lack are structured edges/relations between the actual studies
  • ex:confirms, ex:seemsToConfirm -> ex:confirmatoryEdge
    • [strength of association [“magnitude”]]
  • ex:reproduces, ex:seemsToReproduce -> ex:reproducibilityEdge
  • ex:refutes
  • ex:disproves
  • TODO: see the list i brainstormed -
  • frameworks for edges:
    • NetworkX
    • RDF
      • http://patterns.dataincubator.org/book/qualified-relation.html
      • http://patterns.dataincubator.org/book/nary-relation.html
    • OpenCog AtomSpace (strength of association)

• Develop best practices guidelines and and/or an RDF schema and vocabulary (“repro:) for linking between studies, their supporting data, and their collection methods with URIs.
  • developing vocabularies:
    • Semantic Web Tools
    • Git, GitHub Pages
    • [ ] Schema.org extension vocabularies
  • linked reproduciblity edges:
    • similarTo
    • concursWith
    • discordantWith
    • intendedToReproduce
    • reproduces
  • linked reproducibility classes and properties:
    • [x] schema.org/MedicalStudy, MedicalObservationalStudy, MedicalTrial
      • [ ] https://github.com/twamarc/ScheMed
        • http://schema.org/MedicalTrialDesign
        • http://schema.org/DoubleBlindedTrial
        • http://schema.org/InternationalTrial
        • http://schema.org/MultiCenterTrial
        • http://schema.org/OpenTrial
        • http://schema.org/PlaceboControlledTrial
        • http://schema.org/RandomizedTrial
        • http://schema.org/SingleBlindedTrial
        • http://schema.org/SingleCenterTrial
        • http://schema.org/TripleBlindedTrial
    • See: https://westurner.github.io/opengov/us/#personal-health-agenda

TODO: - pandas 3402 -

StructuredPremises: Premises as structured data

• And then URIs for controls / study design

  • see schema.org/MedicalTrialDesign
    • [ ] these could/should be extended to all of science

• logical premises (sequence of propositions)

• i/o sequences

  • nbformat (IPython / Jupyter notebook format)
    • insufficient because we need stable premise permalinks (across versioned publishing URIs)
      • #premise-1
      • #premise-abc398f

• conclusions (derivations)

  • this is a computation graph
  • it should have links (edges) to the datasets
    • https://schema.org/Dataset
    • “ENH: Linked Datasets (RDF)” https://github.com/pydata/pandas/issues/3402
  • figures should have links (edges) to the datasets
    • permalinks to premises
  • #TenSimpleRules for Reproducibile Computational Research | http://www.ploscompbiol.org/article/info%3Adoi%2F10.1371%2Fjournal.pcbi.1003285 | https://wrdrd.github.io/docs/consulting/data-science#tensimplerules-for-reproducible-computational-research

• further questions

  • “downstream” studies / implementations
  • retraction management
  • decisions / policy predicated on said conclusions

LinkedMetaAnalyses

• You evaluated 10, I evaluated (the same / a different) 10 studies

  • PRISMA

    • Homepage: http://www.prisma-statement.org/

    • http://www.prisma-statement.org/PRISMAStatement/

    • Checklist: http://www.prisma-statement.org/documents/PRISMA%202009%20checklist.pdf

    • Flow Diagram: http://www.prisma-statement.org/documents/PRISMA%202009%20flow%20diagram.pdf

  • evaluation of controls

    • “the URI says they did a Triple Blind Study, but it doesn’t sound like they had groups named just e.g. X, Y, and Z”
      • disqualified / questionable / etc
      • schema.org/MedicalTrial -> schema.org/ScientificTrial

  • C = Class (RDFS)

  • P = Property (RDFS)

  • schema.org/

    • [ ] C: MetaAnalysis
      • [ ] C: CriteriaBase type
        • [ ] C: Criterion
      • [ ] C: ScientificStudy
        • [x] C: MedicalStudy
          • [ ] C: MedicalObservationalStudy <- ScientificObservationalStudy
          • [ ] C: MedicalTrial <- ScientificTrial
      • [x] C: Dataset

• When do we show?

  • Deadline
  • Only if you also produce your own meta-analyses
  • Only if we’re doing Open Access (as required by stipulations of federal funding)

RDF Example

Linked Data + Reproducibility => Linked Reproducibility

Reproducibility ---\___  Linked Reproducibility
Linked Data     ---/

In Turtle RDF syntax:

:LinkedData rdf:type skos:Concept ;
    rdfs:label "Linked Data"@en ;
    schema:name "Linked Data"@en ;
    owl:sameAs <https://en.wikipedia.org/wiki/Linked_data> ;
    owl:sameAs <http://dbpedia.org/page/Linked_data> ;

    owl:sameAs <http://ja.dbpedia.org/resource/Linked_data>
    owl:sameAs <http://es.dbpedia.org/resource/Datos_enlazados> ;
    owl:sameAs <http://fr.dbpedia.org/resource/Web_des_donn%C3%A9es> ;
    owl:sameAs <http://nl.dbpedia.org/resource/Linked_data>
    owl:sameAs <http://ko.dbpedia.org/resource/링크드_데이터> ;
    owl:sameAs <http://wikidata.org/entity/Q515701> ;
    .

:Reproducibility a skos:Concept ;
    rdfs:label "Reproducibility"@en ;
    schema:name "Reproducibility"@en ;
    owl:sameAs <https://en.wikipedia.org/wiki/Reproducibility> ;
    owl:sameAs <http://dbpedia.org/page/Reproducibility> ;
    .

:LinkedReproducibility a skos:Concept ;
    rdfs:label "Linked Reproducibility"@en ;
    schema:name "Linked Reproducibility"@en ;
    skos:related [ :LinkedData, :Reproducibility ] ;
    .

CSV, CSVW, and metadata rows

• CSV – Comma Separated Values
• CSVW – CSV on the Web ( RDF, JSON, JSON-LD )
• RDF – Resource Description Framework
  • A two dimensional spreadsheet containing just data values (e.g. CSV) is a projection of an RDF graph which can be expressed as CSVW.

A classic data table with 1 metadata header row (column label):

Table with 1 metadata header row

column label	sample	date	width	height
	1	2016-06-19T06:28:49-0500	20.0	30.0
	2	2016-06-19T06:29:22-0500	40.0	50.0
	3	2016-06-19T06:29:48-0500	60.0	70.0

A data table with 7 metadata header rows (column label, property URI path, DataType, unit, accuracy, precision, significant figures):

Table with 7 metadata header rows

column label	sample	date	width	height
property URI path	schema.org/name	schema.org/dateCreated	[schema.org/height, schema.org/value]	[schema.org/width, schema.org/value]
schema.org/DataType	schema.org/Integer	schema.org/Date	schema.org/Float	schema.org/Float
Unit			unit:Meter	unit:Meter
accuracy
precision
significant figures			*.1	*.1
	1	2016-06-19T06:28:49-0500	20.0	30.0
	2	2016-06-19T06:29:22-0500	40.0	50.0
	3	2016-06-19T06:29:48-0500	60.0	70.0

• https://en.wikipedia.org/wiki/Significant_figures
• https://en.wikipedia.org/wiki/Accuracy_and_precision
• QUDT unit:
  • http://www.qudt.org/qudt/owl/1.0.0/unit/Instances.html#Meter
  • http://prefix.cc/unit:Meter
  • These should be full URIs ( URIs for Units )

References

• TODO: @westurner
  • reddit
  • twitter
  • hackernews