multidigraph_to_digraph transitive reduction option

[dhimmel]

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

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

[dhimmel]

GO edges removed by reduction

Transitive reduction of GO went from 87,436 to 79,946 by removing 7,490 redundant edges.

Examples of removed edges:

• GO:0002213 --> GO:1900366
  defense response to insect --> negative regulation of defense response to insect
• GO:0016331 --> GO:0021994
  morphogenesis of embryonic epithelium --> progression of neural tube closure
• GO:0033621 --> GO:0120272
  nuclear-transcribed mRNA catabolic process, meiosis-specific transcripts --> positive regulation of nuclear-transcribed mRNA catabolic process, meiosis-specific transcripts
• GO:0034645 --> GO:2000113
  cellular macromolecule biosynthetic process --> negative regulation of cellular macromolecule biosynthetic process
• GO:0035710 --> GO:2000515
  CD4-positive, alpha-beta T cell activation --> negative regulation of CD4-positive, alpha-beta T cell activation
• GO:0043500 --> GO:0014745
  muscle adaptation --> negative regulation of muscle adaptation
• GO:0044839 --> GO:1902750
  cell cycle G2/M phase transition --> negative regulation of cell cycle G2/M phase transition
• GO:0097695 --> GO:1904914
  establishment of protein-containing complex localization to telomere --> negative regulation of establishment of protein-containing complex localization to telomere
• GO:1900868 --> GO:1900972
  sarcinapterin biosynthetic process --> negative regulation of sarcinapterin biosynthetic process
• GO:1990048 --> GO:1901952
  anterograde neuronal dense core vesicle transport --> negative regulation of anterograde dense core granule transport

Source

import random
import networkx as nx
import pronto
from nxontology.imports import pronto_to_multidigraph, multidigraph_to_digraph
url = "http://release.geneontology.org/2021-02-01/ontology/go-basic.json.gz"
go_pronto = pronto.Ontology(handle=url)
go_multidigraph = pronto_to_multidigraph(go_pronto)
go_digraph_full = multidigraph_to_digraph(go_multidigraph, reduce=False)
go_digraph_reduced = multidigraph_to_digraph(go_multidigraph, reduce=True)
removed_edges = sorted(go_digraph_full.edges - go_digraph_reduced.edges)
print(f"Transitive reduction of GO went from {go_digraph_full.number_of_edges():,} to {go_digraph_reduced.number_of_edges():,} by removing {len(removed_edges):,} redundant edges.")
random.seed(0)
sample_removed_edges = sorted(random.sample(removed_edges, k=10))
for source, target in sample_removed_edges:
    print(f"- `{source}` --> `{target}`\n   {go_pronto.get_term(source).name} --> {go_pronto.get_term(target).name}")

[dhimmel]

Visualizing redundant edges

From AmiGo under Graph Views (Graphical View PNG), here are the ancestors (superterms) for GO:1900366 (negative regulation of defense response to insect, first example in last coment)

Notice the red edge from "negative regulation of defense response to insect" to "defense response to insect". This edge (along with other red edges and some black edges) are be removed by the transitive reduction.

[dhimmel]

Hat tip to @eric-czech for first alerting me that read_gene_ontology was producing redundant edges that could be removed by transitive reduction. @eric-czech IIRC you mentioned the reduction step was prohibitively slow. But it's only taking a couple seconds for me on GO. Maybe you were referring to a different ontology? Hence, I think it makes sense to apply the reduction by default to GO.

[eric-czech]

you mentioned the reduction step was prohibitively slow. But it's only taking a couple seconds for me on GO. Maybe you were referring to a different ontology?

Hmm could be, though I thought I had attempted it on GO. I was working with the reverse orientation of the edges so maybe that influences the running time a lot?