multidigraph_to_digraph transitive reduction option (#16)

merges #16

Background information at
https://twitter.com/larsjuhljensen/status/1450188835032375300

Ontologies such as GO can be reduced when collapsing multiple relationship
types into a single relationship type DiGraph.

Improve multidigraph_to_digraph logging

README.md

@@ -124,7 +124,7 @@ Counter({'is a': 71509,
  'regulates': 3216,
  'negatively regulates': 2768,
  'positively regulates': 2756})
->>> go_digraph = multidigraph_to_digraph(go_multidigraph)
+>>> go_digraph = multidigraph_to_digraph(go_multidigraph, reduce=True)
 >>> go_nxo = NXOntology(go_digraph)
 >>> # Notice the similarity increases due to the full set of edges
 >>> round(go_nxo.similarity("GO:0042552", "GO:0022008").lin, 3)

nxontology/imports.py

@@ -142,7 +142,10 @@ def pronto_to_multidigraph(
 
 
 def multidigraph_to_digraph(
- graph: nx.MultiDiGraph, rel_types: Optional[List[str]] = None, reverse: bool = True
+ graph: nx.MultiDiGraph,
+ rel_types: Optional[List[str]] = None,
+ reverse: bool = True,
+ reduce: bool = False,
 ) -> nx.DiGraph:
  """
  Convert a networkx MultiDiGraph to a DiGraph by aggregating edges accross relationship types.
@@ -152,7 +155,13 @@ def multidigraph_to_digraph(
 
  When rel_types is None, all relationship types are preserved. If rel_types is defined,
  then the MultiDiGraph is first filtered for edges with that key (relationship type).
+
+ If reduce is True, perform a transitive reduction on the DiGraph
+ to produce a minimum equivalent graph that removes redundant relationships
+ — i.e. those that are already captured by a more specific ancestral path.
+ The default is reduce=False since the reduction can be a computationally expensive step.
  """
+ logging.info(f"Received MultiDiGraph with {graph.number_of_edges():,} edges.")
  if rel_types is not None:
  graph.remove_edges_from(
  [
@@ -161,11 +170,23 @@ def multidigraph_to_digraph(
  if key not in rel_types
  ]
  )
+ logging.info(
+ f"Filtered MultiDiGraph to {graph.number_of_edges():,} edges of the following types: {rel_types}."
+ )
  if reverse:
  graph = graph.reverse(copy=True)
  digraph = nx.DiGraph(graph)
+ if reduce:
+ n_edges_before = digraph.number_of_edges()
+ digraph = nx.transitive_reduction(digraph)
+ logging.info(
+ f"Reduced DiGraph by removing {n_edges_before - digraph.number_of_edges():,} redundant edges."
+ )
  for source, target in digraph.edges(data=False):
  digraph[source][target]["rel_types"] = sorted(graph[source][target])
+ logging.info(
+ f"Converted MultiDiGraph to DiGraph with {digraph.number_of_nodes():,} nodes and {digraph.number_of_edges():,} edges."
+ )
  return digraph
 
 
@@ -179,6 +200,7 @@ def read_gene_ontology(
  "negatively regulates",
  "positively regulates",
  ],
+ reduce: bool = True,
 ) -> NXOntology[str]:
  """
  Load the Gene Ontology into NXOntology,
@@ -198,10 +220,14 @@ def read_gene_ontology(
  else:
  date.fromisoformat(release) # check that release is a valid date
  url = f"http://release.geneontology.org/{release}/ontology/{source_file}"
+ logging.info(f"Loading Gene Ontology into Pronto from <{url}>.")
  go_pronto = Prontology(handle=url)
  go_multidigraph = pronto_to_multidigraph(go_pronto, default_rel_type="is a")
  go_digraph = multidigraph_to_digraph(
- go_multidigraph, rel_types=rel_types, reverse=True
+ go_multidigraph,
+ rel_types=rel_types,
+ reverse=True,
+ reduce=reduce,
  )
  go_nxo: NXOntology[str] = NXOntology(go_digraph)
  go_nxo.graph.graph["source_url"] = url

nxontology/tests/imports_test.py

@@ -90,3 +90,11 @@ def test_read_gene_ontology():
  == "http://release.geneontology.org/2021-02-01/ontology/go-basic.json.gz"
  )
  assert "regulates" in nxo.graph["GO:0006310"]["GO:0000018"]["rel_types"]
+ # Transitive reduction should remove this edge
+ # from "defense response to insect" to "negative regulation of defense response to insect"
+ # since it is redundant with a more specific ancestral path.
+ # https://github.com/related-sciences/nxontology/pull/16
+ assert not nxo.graph.has_edge("GO:0002213", "GO:1900366")
+ # GO:0002213 --> GO:2000068 --> GO:1900366 is more specific
+ assert nxo.graph.has_edge("GO:0002213", "GO:2000068")
+ assert nxo.graph.has_edge("GO:2000068", "GO:1900366")