Annotations¶

A model element on its own carries no meaning a machine can use. A species named glc could be glucose, a glucose transporter or a parameter someone abbreviated. MIRIAM annotations solve this by attaching two things to an element:

• a qualifier saying how the element relates to something else, e.g., "is", "is part of", "is version of"
• a resource pointing at a database entry, e.g., CHEBI:17234

Together they form a statement: this species is the chemical entity CHEBI:17234. pymetadata provides the qualifiers, parses resources written in any of the common notations, normalizes them, and validates them against the identifiers.org registry.

Qualifiers¶

Biological qualifiers (BQB) relate an element to a biological entity, model qualifiers (BQM) describe the model itself:

from pymetadata.core.miriam import BQB, BQM

BQB.IS  # the element is the annotated entity
BQB.IS_VERSION_OF  # the element is a version of the entity
BQB.IS_PART_OF  # the element is part of the entity
BQB.HAS_TAXON  # the entity occurs in the given taxon
BQM.IS_DESCRIBED_BY  # the model is described by the resource, e.g., a publication

Choosing the right qualifier matters: BQB.IS on a species that is only one form of a compound is a stronger claim than the model supports, which is what BQB.IS_VERSION_OF is for.

Resources¶

RDFAnnotation accepts the notations found in the wild and reduces them to a collection and a term:

from pymetadata.core.annotation import RDFAnnotation
from pymetadata.core.miriam import BQB

for resource in [
    "CHEBI:33699",  # compact identifier
    "chebi/CHEBI:33699",  # collection and term
    "https://identifiers.org/CHEBI:33699",  # identifiers.org URL
    "http://identifiers.org/chebi/CHEBI:33699",  # legacy identifiers.org URL
    "urn:miriam:chebi:CHEBI%3A33699",  # deprecated MIRIAM URN
]:
    annotation = RDFAnnotation(qualifier=BQB.IS, resource=resource)
    print(annotation.resource_normalized)
# https://identifiers.org/CHEBI:33699

All five spellings normalize to the same compact identifier. Arbitrary URLs are also valid resources; they are kept as they are, because there is no registry entry to normalize them against:

RDFAnnotation(qualifier=BQB.IS, resource="https://en.wikipedia.org/wiki/Cytosol")

resource_normalized returns https://identifiers.org/<prefix>:<accession>. For namespaces which embed their prefix in the identifier (GO, CHEBI, SBO, BTO) the prefix is already part of the term and is not added a second time.

Validation¶

Validation answers two questions: is the qualifier a real MIRIAM qualifier, and does the term match the pattern the identifiers.org registry defines for its collection?

annotation = RDFAnnotation(qualifier=BQB.IS, resource="chebi/CHEBI:33699")
annotation.validate()  # qualifier and term
annotation.check_miriam_term()  # term against the registry pattern only

A term such as chebi/CHEBI:X33699 fails, because the CHEBI pattern is ^CHEBI:\d+$. The registry is downloaded once and cached, see Installation.

Resolving additional information¶

RDFAnnotationData takes an annotation and resolves what the identifier actually refers to.

Constructing it resolves the cross references: for every provider the identifiers.org registry lists for the collection, the provider's URL pattern is filled in with the term, giving one CrossReference per provider.

from pymetadata.core.annotation import RDFAnnotation, RDFAnnotationData
from pymetadata.core.miriam import BQB

annotation = RDFAnnotation(qualifier=BQB.IS, resource="chebi/CHEBI:33699")
data = RDFAnnotationData(annotation)
print(data.url)  # url of the first provider
print(data.xrefs)  # one entry per identifiers.org provider

query_ols() then asks the Ontology Lookup Service for the term itself and fills in label, description and synonyms. Note that it also replaces xrefs with the cross references reported by OLS, which can be empty for a given term:

data.query_ols()

print(data.label)  # messenger RNA
print(data.description)  # An RNA molecule that transfers the coding information ...
print(data.synonyms)  # mRNA, ...

Both steps require network access. Responses are cached on disk by default, so repeated terms are not queried again; see Cache.

Ontology terms¶

Ontology terms are usually passed around as strings, which means typos surface at runtime or not at all, and nothing tells you what a term means. pymetadata ships three ontologies as python classes generated from the ontology releases themselves:

Every term is available under both its identifier and its name:

from pymetadata.ontologies import SBO, PBPKO

SBO.SBO_0000247  # by id
SBO.SIMPLE_CHEMICAL  # by name
PBPKO.BODYWEIGHT

SBO.get_name(SBO.SIMPLE_CHEMICAL)
# 'simple chemical'

validate accepts both the underscore and the colon notation and returns the enum member, which is the convenient way to accept terms from user input or from a file:

SBO.validate("SBO:0000247")  # SBO.SBO_0000247
SBO.validate("SBO_0000247")  # SBO.SBO_0000247
SBO.validate("SBO_9999999")  # raises AttributeError, the term does not exist

What a term knows¶

A term is the identifier, i.e., a str, and carries what the ontology release says about it. This is the information which otherwise has to be looked up in OLS, and it is available offline. Since every term is a documented attribute of its ontology, an editor completes SBO. and shows the definition of the term it suggests:

term = SBO.SIMPLE_CHEMICAL

term.value  # 'SBO_0000247', the enum is a str subclass
term.label  # 'simple chemical'
term.definition  # 'Simple, non-repetitive chemical entity.'
term.synonyms  # ()
term.deprecated  # False, the term is not obsolete
term.curie  # 'SBO:0000247'
term.url  # 'https://identifiers.org/SBO:0000247'

repr(term)
# <SBO.SBO_0000247: 'simple chemical'>

get_term resolves a term given as a string, and the ontology behaves like the enum it replaces, i.e., it can be iterated, asked for its length and indexed:

from pymetadata.ontologies import KISAO

KISAO.get_term("KISAO:0000019").synonyms
# ('VODE', 'VODEPK', 'code value ordinary differential equation solver')

len(KISAO)  # number of terms
KISAO["KISAO_0000019"]  # lookup by identifier
"KISAO:0000019" in KISAO  # True
[term for term in KISAO if term.deprecated]

A term stays a string, i.e., SBO.SIMPLE_CHEMICAL == "SBO_0000247" is true and serializing a term gives the identifier, so terms can be passed wherever the identifier is expected. The name in the ontology is label, name is the identifier.

The ontologies are generated with the internal pymetadata.ontologies._ontology_builder, which downloads the ontology in OWL format and writes a python module from it. See Development for how to update them to a newer ontology release.