import networkx as nx
from typing import List, Set, Dict, Tuple, Union, Optional, Any
import rdflib
from biolinkml.meta import SlotDefinition, ClassDefinition, Element
from rdflib import URIRef, Namespace
from kgx.config import get_logger
from kgx.curie_lookup_service import CurieLookupService
from kgx.utils.graph_utils import curie_lookup
from kgx.utils.rdf_utils import property_mapping, is_property_multivalued, reverse_property_mapping
from kgx.utils.kgx_utils import generate_edge_key, get_toolkit, sentencecase_to_camelcase, sentencecase_to_snakecase, \
get_biolink_ancestors
from kgx.prefix_manager import PrefixManager
log = get_logger()
[docs]class RdfGraphMixin(object):
"""
A mixin that defines the following methods,
- load_networkx_graph(): template method that all deriving classes should implement
- add_node(): method to add a node from a RDF form to property graph form
- add_node_attribute(): method to add a node attribute from a RDF form to property graph form
- add_edge(): method to add an edge from a RDF form to property graph form
- add_edge_attribute(): method to add an edge attribute from an RDF form to property graph form
"""
DEFAULT_EDGE_LABEL = 'biolink:related_to'
CORE_NODE_PROPERTIES = {'id'}
CORE_EDGE_PROPERTIES = {'id', 'subject', 'edge_label', 'object', 'type'}
def __init__(self, source_graph: Optional[nx.MultiDiGraph] = None, curie_map: Optional[Dict] = None):
if source_graph:
self.graph = source_graph
else:
self.graph = nx.MultiDiGraph()
self.graph_metadata: Dict = {}
self.prefix_manager = PrefixManager()
self.DEFAULT = Namespace(self.prefix_manager.prefix_map[''])
if curie_map:
self.prefix_manager.update_prefix_map(curie_map)
# TODO: use OBO IRI from biolink model context once https://github.com/biolink/biolink-model/issues/211 is resolved
self.OBO = Namespace('http://purl.obolibrary.org/obo/')
self.OBAN = Namespace(self.prefix_manager.prefix_map['OBAN'])
self.PMID = Namespace(self.prefix_manager.prefix_map['PMID'])
self.BIOLINK = Namespace(self.prefix_manager.prefix_map['biolink'])
self.predicate_mapping = property_mapping.copy()
self.reverse_predicate_mapping = reverse_property_mapping.copy()
self.cache: Dict = {}
[docs] def load_networkx_graph(self, rdfgraph: rdflib.Graph, predicates: Optional[Set[URIRef]] = None, **kwargs: Dict) -> None:
"""
This method should be overridden and be implemented by the derived class,
and should load all desired nodes and edges from rdflib.Graph into networkx.MultiDiGraph
Its preferred that this method does not use the networkx API directly
when adding nodes, edges, and their attributes.
Instead, Using the following methods,
- ``add_node()``
- ``add_node_attribute()``
- ``add_edge()``
- ``add_edge_attribute()``
to ensure that nodes, edges, and their attributes
are added in conformance with the BioLink Model, and that URIRef's are
translated into CURIEs or BioLink Model elements whenever appropriate.
Parameters
----------
rdfgraph: rdflib.Graph
Graph containing nodes and edges
predicates: Optional[Set[URIRef]]
A set containing predicates in rdflib.URIRef form
kwargs: Dict
Any additional arguments
"""
raise NotImplementedError("Method not implemented.")
[docs] def add_node(self, iri: URIRef, data: Optional[Dict] = None) -> Dict:
"""
This method should be used by all derived classes when adding a node to
the networkx.MultiDiGraph. This ensures that a node's identifier is a CURIE,
and that it's `iri` property is set.
Returns the CURIE identifier for the node in the networkx.MultiDiGraph
Parameters
----------
iri: rdflib.URIRef
IRI of a node
data: Optional[Dict]
Additional node properties
Returns
-------
Dict
The node data
"""
n = self.prefix_manager.contract(str(iri))
if n == iri:
if self.prefix_manager.has_urlfragment(iri):
n = rdflib.namespace.urldefrag(iri).fragment
if not n:
n = iri
if self.graph.has_node(n):
node_data = self.update_node(n, data)
else:
if data:
node_data = data
else:
node_data = {}
node_data['id'] = n
if 'category' in node_data:
if 'biolink:NamedThing' not in set(node_data['category']):
node_data['category'].append('biolink:NamedThing')
else:
node_data['category'] = ["biolink:NamedThing"]
if 'provided_by' in self.graph_metadata and 'provided_by' not in node_data:
node_data['provided_by'] = self.graph_metadata['provided_by']
self.graph.add_node(n, **node_data)
return node_data
[docs] def update_node(self, n: Union[URIRef, str], data: Optional[Dict] = None) -> Dict:
"""
Update a node with properties.
Parameters
----------
n: Union[URIRef, str]
Node identifier
data: Optional[Dict]
Node properties
Returns
-------
Dict
The node data
"""
node_data = self.graph.nodes[str(n)]
if data:
new_data = self._prepare_data_dict(node_data, data)
node_data.update(new_data)
return node_data
[docs] def add_edge(self, subject_iri: URIRef, object_iri: URIRef, predicate_iri: URIRef, data: Optional[Dict[Any, Any]] = None) -> Dict:
"""
This method should be used by all derived classes when adding an edge to the networkx.MultiDiGraph.
This method ensures that the `subject` and `object` identifiers are CURIEs, and that `edge_label`
is in the correct form.
Parameters
----------
subject_iri: rdflib.URIRef
Subject IRI for the subject in a triple
object_iri: rdflib.URIRef
Object IRI for the object in a triple
predicate_iri: rdflib.URIRef
Predicate IRI for the predicate in a triple
data: Optional[Dict[Any, Any]]
Additional edge properties
Returns
-------
Dict
The edge data
"""
(element_uri, canonical_uri, predicate, property_name) = self.process_predicate(predicate_iri)
subject_node = self.add_node(subject_iri)
object_node = self.add_node(object_iri)
edge_label = element_uri if element_uri else predicate
if not edge_label:
edge_label = property_name
if ' ' in edge_label:
log.debug(f"predicate IRI '{predicate_iri}' yields edge_label '{edge_label}' that not in snake_case form; replacing ' ' with '_'")
edge_label_prefix = self.prefix_manager.get_prefix(edge_label)
if edge_label_prefix not in {'biolink', 'rdf', 'rdfs', 'skos', 'owl'}:
if PrefixManager.is_curie(edge_label):
# name = curie_lookup(edge_label)
# if name:
# log.debug(f"predicate IRI '{predicate_iri}' yields edge_label '{edge_label}' that is actually a CURIE; Using its mapping instead: {name}")
# edge_label = f"{edge_label_prefix}:{name}"
# else:
# log.debug(f"predicate IRI '{predicate_iri}' yields edge_label '{edge_label}' that is actually a CURIE; defaulting back to {self.DEFAULT_EDGE_LABEL}")
edge_label = self.DEFAULT_EDGE_LABEL
edge_key = generate_edge_key(subject_node['id'], edge_label, object_node['id'])
if self.graph.has_edge(subject_node['id'], object_node['id'], key=edge_key):
# edge already exists; process kwargs and update the edge
edge_data = self.update_edge(subject_node['id'], object_node['id'], edge_key, data)
else:
# add a new edge
edge_data = data if data else {}
edge_data.update({
'subject': subject_node['id'],
'edge_label': f"{edge_label}",
'object': object_node['id']
})
if 'relation' not in edge_data:
edge_data['relation'] = predicate
if 'provided_by' in self.graph_metadata and 'provided_by' not in edge_data:
edge_data['provided_by'] = self.graph_metadata['provided_by']
self.graph.add_edge(subject_node['id'], object_node['id'], key=edge_key, **edge_data)
return edge_data
[docs] def update_edge(self, subject_curie: str, object_curie: str, edge_key: str, data: Optional[Dict[Any, Any]]) -> Dict:
"""
Update an edge with properties.
Parameters
----------
subject_curie: str
Subject CURIE
object_curie: str
Object CURIE
edge_key: str
Edge key
data: Optional[Dict[Any, Any]]
Edge properties
Returns
-------
Dict
The edge data
"""
edge_data = self.graph.get_edge_data(subject_curie, object_curie, key=edge_key)
if data:
new_data = self._prepare_data_dict(edge_data, data)
edge_data.update(new_data)
return edge_data
[docs] def add_node_attribute(self, iri: Union[URIRef, str], key: str, value: Union[str, List]) -> Dict:
"""
Add an attribute to a node, while taking into account whether the attribute
should be multi-valued.
Multi-valued properties will not contain duplicates.
The ``key`` may be a rdflib.URIRef or a URI string that maps onto a property name
as defined in ``rdf_utils.property_mapping``.
If the node does not exist then it is created.
Parameters
----------
iri: Union[rdflib.URIRef, str]
The IRI of a node in the rdflib.Graph
key: str
The name of the attribute. Can be a rdflib.URIRef or URI string
value: Union[str, List]
The value of the attribute
Returns
-------
Dict
The node data
"""
if self.prefix_manager.is_iri(key):
key_curie = self.prefix_manager.contract(key)
else:
key_curie = key
c = curie_lookup(key_curie)
if c:
key_curie = c
if self.prefix_manager.is_curie(key_curie):
# property names will always be just the reference
mapped_key = self.prefix_manager.get_reference(key_curie)
else:
mapped_key = key_curie
if isinstance(value, rdflib.term.Identifier):
if isinstance(value, rdflib.term.URIRef):
value_curie = self.prefix_manager.contract(value)
# if self.prefix_manager.get_prefix(value_curie) not in {'biolink'} \
# and mapped_key not in {'type', 'category', 'edge_label', 'relation', 'predicate'}:
# d = self.add_node(value)
# value = d['id']
# else:
# value = value_curie
value = value_curie
else:
value = value.toPython()
if mapped_key in is_property_multivalued and is_property_multivalued[mapped_key]:
value = [value]
node_data = self.add_node(iri, {mapped_key: value})
return node_data
[docs] def add_edge_attribute(self, subject_iri: Union[URIRef, str], object_iri: URIRef, predicate_iri: URIRef, key: str, value: str) -> Dict:
"""
Adds an attribute to an edge, while taking into account whether the attribute
should be multi-valued.
Multi-valued properties will not contain duplicates.
The ``key`` may be a rdflib.URIRef or a URI string that maps onto a property name
as defined in ``rdf_utils.property_mapping``.
If the nodes in the edge does not exist then they will be created.
Parameters
----------
subject_iri: [rdflib.URIRef, str]
The IRI of the subject node of an edge in rdflib.Graph
object_iri: rdflib.URIRef
The IRI of the object node of an edge in rdflib.Graph
predicate_iri: rdflib.URIRef
The IRI of the predicate representing an edge in rdflib.Graph
key: str
The name of the attribute. Can be a rdflib.URIRef or URI string
value: str
The value of the attribute
Returns
-------
Dict
The edge data
"""
pass
def _prepare_data_dict(self, d1: Dict, d2: Dict) -> Dict:
"""
Given two dict objects, make a new dict object that is the intersection of the two.
If a key is known to be multivalued then it's value is converted to a list.
If a key is already multivalued then it is updated with new values.
If a key is single valued, and a new unique value is found then the existing value is
converted to a list and the new value is appended to this list.
Parameters
----------
d1: Dict
Dict object
d2: Dict
Dict object
Returns
-------
Dict
The intersection of d1 and d2
"""
new_data = {}
for key, value in d2.items():
if isinstance(value, (list, set, tuple)):
new_value = [self.prefix_manager.contract(x) if self.prefix_manager.is_iri(x) else x for x in value]
else:
new_value = self.prefix_manager.contract(value) if self.prefix_manager.is_iri(value) else value
if key in is_property_multivalued:
if is_property_multivalued[key]:
# key is supposed to be multivalued
if key in d1:
# key is in data
if isinstance(d1[key], list):
# existing key has value type list
new_data[key] = d1[key]
if isinstance(new_value, (list, set, tuple)):
new_data[key] += [x for x in new_value if x not in new_data[key]]
else:
if new_value not in new_data[key]:
new_data[key].append(new_value)
else:
if key in self.CORE_NODE_PROPERTIES or key in self.CORE_EDGE_PROPERTIES:
log.debug(f"cannot modify core property '{key}': {d2[key]} vs {d1[key]}")
else:
# existing key does not have value type list; converting to list
new_data[key] = [d1[key]]
if isinstance(new_value, (list, set, tuple)):
new_data[key] += [x for x in new_value if x not in new_data[key]]
else:
if new_value not in new_data[key]:
new_data[key].append(new_value)
else:
# key is not in data; adding
if isinstance(new_value, (list, set, tuple)):
new_data[key] = [x for x in new_value]
else:
new_data[key] = [new_value]
else:
# key is not multivalued; adding/replacing as-is
if key in d1:
if isinstance(d1[key], list):
new_data[key] = d1[key]
if isinstance(new_value, (list, set, tuple)):
new_data[key] += [x for x in new_value]
else:
new_data[key].append(new_value)
else:
if key in self.CORE_NODE_PROPERTIES or key in self.CORE_EDGE_PROPERTIES:
log.debug(f"cannot modify core property '{key}': {d2[key]} vs {d1[key]}")
else:
new_data[key] = new_value
else:
new_data[key] = new_value
else:
# treating key as multivalued
if key in d1:
# key is in data
if key in self.CORE_NODE_PROPERTIES or key in self.CORE_EDGE_PROPERTIES:
log.debug(f"cannot modify core property '{key}': {d2[key]} vs {d1[key]}")
else:
if isinstance(d1[key], list):
# existing key has value type list
new_data[key] = d1[key]
if isinstance(new_value, (list, set, tuple)):
new_data[key] += [x for x in new_value if x not in new_data[key]]
else:
new_data[key].append(new_value)
else:
# existing key does not have value type list; converting to list
new_data[key] = [d1[key]]
if isinstance(new_value, (list, set, tuple)):
new_data[key] += [x for x in new_value if x not in new_data[key]]
else:
new_data[key].append(new_value)
else:
new_data[key] = new_value
return new_data
[docs] def get_biolink_element(self, predicate: Any) -> Optional[Element]:
"""
Returns a Biolink Model element for a given predicate.
Parameters
----------
predicate: Any
The CURIE of a predicate
Returns
-------
Optional[Element]
The corresponding Biolink Model element
"""
toolkit = get_toolkit()
if self.prefix_manager.is_iri(predicate):
predicate_curie = self.prefix_manager.contract(predicate)
else:
predicate_curie = predicate
if self.prefix_manager.is_curie(predicate_curie):
reference = self.prefix_manager.get_reference(predicate_curie)
else:
reference = predicate_curie
element = toolkit.get_element(reference)
if not element:
try:
mapping = toolkit.get_by_mapping(predicate)
element = toolkit.get_element(mapping)
except ValueError as e:
log.error(e)
return element
[docs] def process_predicate(self, p: Optional[Union[URIRef, str]]) -> Tuple[str, str, str, str]:
"""
Process a predicate where the method checks if there is a mapping in Biolink Model.
Parameters
----------
p: Optional[Union[URIRef, str]]
The predicate
Returns
-------
Tuple[str, str, str, str]
A tuple that contains the Biolink CURIE (if available), the Biolink slot_uri CURIE (if available),
the CURIE form of p, the reference of p
"""
if p in self.cache:
# already processed this predicate before; pull from cache
element_uri = self.cache[p]['element_uri']
canonical_uri = self.cache[p]['canonical_uri']
predicate = self.cache[p]['predicate']
property_name = self.cache[p]['property_name']
else:
# haven't seen this property before; map to element
if self.prefix_manager.is_iri(p):
predicate = self.prefix_manager.contract(str(p))
else:
predicate = None
if self.prefix_manager.is_curie(p):
property_name = self.prefix_manager.get_reference(p)
predicate = p
else:
if predicate and self.prefix_manager.is_curie(predicate):
property_name = self.prefix_manager.get_reference(predicate)
else:
property_name = p
predicate = f":{p}"
element = self.get_biolink_element(p)
canonical_uri = None
if element:
if isinstance(element, SlotDefinition):
# predicate corresponds to a biolink slot
if element.definition_uri:
element_uri = self.prefix_manager.contract(element.definition_uri)
else:
element_uri = f"biolink:{sentencecase_to_snakecase(element.name)}"
if element.slot_uri:
canonical_uri = element.slot_uri
elif isinstance(element, ClassDefinition):
# this will happen only when the IRI is actually
# a reference to a class
element_uri = self.prefix_manager.contract(element.class_uri)
else:
element_uri = f"biolink:{sentencecase_to_camelcase(element.name)}"
if 'biolink:Attribute' in get_biolink_ancestors(element.name):
element_uri = f"biolink:{sentencecase_to_snakecase(element.name)}"
if not predicate:
predicate = element_uri
else:
# no mapping to biolink model;
# look at predicate mappings
element_uri = None
if p in self.predicate_mapping:
property_name = self.predicate_mapping[p]
predicate = f":{property_name}"
self.cache[p] = {'element_uri': element_uri, 'canonical_uri': canonical_uri, 'predicate': predicate, 'property_name': property_name}
return element_uri, canonical_uri, predicate, property_name