diff options
author | Munyoki Kilyungi | 2024-09-11 11:50:53 +0300 |
---|---|---|
committer | John Nduli Kilyungi | 2024-09-18 08:32:34 +0300 |
commit | 679bfa89863bae7ea89426e0e0e15e746d4b36fa (patch) | |
tree | 76e153b91fd8b25389fa2674d7dd163919eaa642 /gn3 | |
parent | 7842de892b07c1df72bf186ff9efad7eb6955171 (diff) | |
download | genenetwork3-679bfa89863bae7ea89426e0e0e15e746d4b36fa.tar.gz |
Run python-black against file.
Signed-off-by: Munyoki Kilyungi <me@bonfacemunyoki.com>
Diffstat (limited to 'gn3')
-rw-r--r-- | gn3/db/rdf/__init__.py | 73 |
1 files changed, 37 insertions, 36 deletions
diff --git a/gn3/db/rdf/__init__.py b/gn3/db/rdf/__init__.py index ffb75e5..6984497 100644 --- a/gn3/db/rdf/__init__.py +++ b/gn3/db/rdf/__init__.py @@ -35,8 +35,9 @@ PREFIXES = { "xsd": "http://www.w3.org/2001/XMLSchema#", } -RDF_PREFIXES = "\n".join([f"PREFIX {key}: <{value}>" - for key, value in PREFIXES.items()]) +RDF_PREFIXES = "\n".join( + [f"PREFIX {key}: <{value}>" for key, value in PREFIXES.items()] +) BASE_CONTEXT = { "data": "@graph", @@ -126,39 +127,40 @@ PUBLICATION_CONTEXT = { "@id": "fabio:hasPublicationYear", "@type": "xsd:gYear", }, - "month": { - "@id": "prism:publicationDate", - "@type": "xsd:gMonth" - }, + "month": {"@id": "prism:publicationDate", "@type": "xsd:gMonth"}, } -PHENOTYPE_CONTEXT = BASE_CONTEXT | PUBLICATION_CONTEXT | { - "skos": "http://www.w3.org/2004/02/skos/core#", - "dcat": "http://www.w3.org/ns/dcat#", - "prism": "http://prismstandard.org/namespaces/basic/2.0/", - "traitName": "skos:altLabel", - "trait": "rdfs:label", - "altName": "rdfs:altLabel", - "description": "dct:description", - "abbreviation": "gnt:abbreviation", - "labCode": "gnt:labCode", - "submitter": "gnt:submitter", - "dataset": "dcat:Distribution", - "contributor": "dct:contributor", - "mean": "gnt:mean", - "locus": "gnt:locus", - "lodScore": "gnt:lodScore", - "references": "dct:isReferencedBy", - "additive": "gnt:additive", - "sequence": "gnt:sequence", - "prefLabel": "skos:prefLabel", - "identifier": "dct:identifier", - "chromosome": "gnt:chr", - "mb": "gnt:mb", - "peakLocation": "gnt:locus", - "species": "gnt:belongsToSpecies", - "group": "gnt:belongsToGroup", -} +PHENOTYPE_CONTEXT = ( + BASE_CONTEXT + | PUBLICATION_CONTEXT + | { + "skos": "http://www.w3.org/2004/02/skos/core#", + "dcat": "http://www.w3.org/ns/dcat#", + "prism": "http://prismstandard.org/namespaces/basic/2.0/", + "traitName": "skos:altLabel", + "trait": "rdfs:label", + "altName": "rdfs:altLabel", + "description": "dct:description", + "abbreviation": "gnt:abbreviation", + "labCode": "gnt:labCode", + "submitter": "gnt:submitter", + "dataset": "dcat:Distribution", + "contributor": "dct:contributor", + "mean": "gnt:mean", + "locus": "gnt:locus", + "lodScore": "gnt:lodScore", + "references": "dct:isReferencedBy", + "additive": "gnt:additive", + "sequence": "gnt:sequence", + "prefLabel": "skos:prefLabel", + "identifier": "dct:identifier", + "chromosome": "gnt:chr", + "mb": "gnt:mb", + "peakLocation": "gnt:locus", + "species": "gnt:belongsToSpecies", + "group": "gnt:belongsToGroup", + } +) def sparql_construct_query(query: str, endpoint: str) -> dict: @@ -174,9 +176,8 @@ def query_frame_and_compact(query: str, context: dict, endpoint: str) -> dict: """Frame and then compact the results given a context""" results = sparql_construct_query(query, endpoint) return jsonld.compact( - jsonld.frame(results, context), - context, - options={"graph": True}) + jsonld.frame(results, context), context, options={"graph": True} + ) def query_and_compact(query: str, context: dict, endpoint: str) -> dict: |