Basic Concepts¶
Ontologies¶
We leverage networkx
Class: Ontology
from ontobio.ontol_factory import OntologyFactory
ont = OntologyFactory().create("go")
[nucleus] = ont.search('nucleus')
ancestors = ont.ancestors(nucleus)
Alternatives¶
Ontobio is aimed primarily at bioinformatics applications, which typically have lightweight ontology requirements: navigation and grouping via graph structures, access to basic metadata such as synonyms.
Associations¶
The association model is a generalization of the GO association/annotation model. The typical scenario is to link a biological entity (gene, gene product, protein complex, variant or allele, disease, individual organism) to a descriptive ontology class, via a defined relationship type, plus metadata such as provenance and evidence. Note that it can be generalized further also link two entities (e.g. gene-gene, such as homology or relationship) or two ontology classes. In fact the distinction between entities and ontology nodes is one of convenience.
Categories¶
TODO
Lightweight vs Complete¶
For many purposes, it is only necessary to use a very lightweight
representation of associations, as a collection of pairwise mappings
between subjects and objects. This can be found in the class AssociationSet. An association set can be constructed using a
particular set of criteria - e.g. all GO annotations to all zebrafish
genes.
For other purposes it is necessary to have a full-blown representation, in which each association is modeled complete with evidence, provenance and so on. TODO Link to documentation.
Example Asssociation Set¶
This example shows a simple set of pairwise associations:
from ontobio.assoc_factory import AssociationSetFactory
afactory = AssociationSetFactory()
aset = afactory.create(ontology=ont,
subject_category='gene',
object_category='function',
taxon='NCBITaxon:7955') ## Zebrafish
Assocations vs ontology edges¶
The distinction between an association (aka annotation) and an ontology edge is primarily one of convenience. For example, it is possible to combine diseases, phenotypes and the associations between them in one graph, with relationship type has-phenotype connecting these. Similary, gene could be added to a GO molecular function graph, connecting via capable-of.
By stratifying the two sets of entities and using a different data structure to connect these, we make it easier to define and perform certain operations, e.g. enrichment, semantic similarity, machine learning, etc.
But we also provide means of interconverting between these two perspectives (TODO).
Identifiers¶
Ontobio uses CURIEs to identify entities, e.g. OMIM:123, GO:0001850. See Identifiers for more information