Using custom networks

In addition to using the mantra database as a basis for enrichment analysis, it is also possible to use self-generated networks. The only restriction for these networks is that they need to have the same node- and edge-types annotated with attributes named ‘node_type’ and ‘edge_type’.

We recommend to also set the additional attributes for compatibility with the provided plotting functions:

• reaction: Formula, Description

• metabolite: Name

• organism: nodeLabel

• gene: nodeLabel

The following names are currently used:

• nodes

  • reaction

  • metabolite

  • organism

  • gene

• edges

  • SUBSTRATE (metabolite - reaction)

  • PRODUCT (reaction - metabolite)

  • REACTION_ORGANISM (reaction - organism)

  • REACTION_GENE (reaction - gene)

We start with a set of metabolite and reaction nodes (and optionally organism and gene nodes) and edges.

import networkx as nx
import matplotlib.pyplot as plt

from pymantra.statics import EDGE_BY_NODE_TYPE
from pymantra.database import reduce_reaction_nodes
from pymantra.plotting import plot_directed_graph

metabolite_nodes = {f"m{i}" for i in range(6)}
reaction_nodes = {f"r{i}" for i in range(3)}
organism_nodes = {f"o{i}" for i in range(4)}

edges = {
    # metabolite - reaction/ reaction - metabolite edges
    ("m0", "r0"), ("r0", "m1"), ("r0", "m2"),
    ("m1", "r1"), ("r1", "m3"), ("r1", "m4"),
    ("m2", "r2"), ("r2", "m5"),
    # organism - reaction edges
    ("o0", "r0"), ("o1", "r0"),
    ("o0", "r1"), ("o2", "r1"), ("o3", "r1"),
    ("o1", "r2"), ("o2", "r2"), ("o3", "r2")
}

The first step is to add the nodes with the required ‘node_type’ attribute.

graph = nx.DiGraph()

for node in metabolite_nodes:
    graph.add_node(node, node_type="metabolite")
for node in reaction_nodes:
    graph.add_node(node, node_type="reaction")
for node in organism_nodes:
    graph.add_node(node, node_type="organism")

Subsequently, the edges can be added and the edge type can directly be inferred using the static variable EDGE_BY_NODE_TYPE, which is a dictionary, where the keys are 2-tuples of source and target node types.

To avoid having multiple reactions with the same substrates and products (these will always get the same reaction activity change), we use the reduce_reaction_nodes function.

for src, tgt in edges:
    src_type = graph.nodes[src]["node_type"]
    tgt_type = graph.nodes[tgt]["node_type"]
    graph.add_edge(
        src, tgt, edge_type=EDGE_BY_NODE_TYPE[(src_type, tgt_type)])

graph = reduce_reaction_nodes(graph)

Finally, we can plot the generated graph. This is also a partial sanity check, if node colours and shapes need to match the match the ones in the legend.

for src, tgt in edges:

The graph can then be used to compute reaction models and multi-omics associations and to run a local search as in Finding dysregulated metabolic reactions and Reaction-based Metabolome-Microbiome Integration. The only requirement is of course that the metabolite, organsim and gene names match between graph and data.

Full Example Code

import networkx as nx
import matplotlib.pyplot as plt

from pymantra.statics import EDGE_BY_NODE_TYPE
from pymantra.database import reduce_reaction_nodes
from pymantra.plotting import plot_directed_graph

metabolite_nodes = {f"m{i}" for i in range(6)}
reaction_nodes = {f"r{i}" for i in range(3)}
organism_nodes = {f"o{i}" for i in range(4)}

edges = {
    # metabolite - reaction/ reaction - metabolite edges
    ("m0", "r0"), ("r0", "m1"), ("r0", "m2"),
    ("m1", "r1"), ("r1", "m3"), ("r1", "m4"),
    ("m2", "r2"), ("r2", "m5"),
    # organism - reaction edges
    ("o0", "r0"), ("o1", "r0"),
    ("o0", "r1"), ("o2", "r1"), ("o3", "r1"),
    ("o1", "r2"), ("o2", "r2"), ("o3", "r2")
}

graph = nx.DiGraph()

for node in metabolite_nodes:
    graph.add_node(node, node_type="metabolite")
for node in reaction_nodes:
    graph.add_node(node, node_type="reaction")
for node in organism_nodes:
    graph.add_node(node, node_type="organism")

for src, tgt in edges:
    src_type = graph.nodes[src]["node_type"]
    tgt_type = graph.nodes[tgt]["node_type"]
    graph.add_edge(
        src, tgt, edge_type=EDGE_BY_NODE_TYPE[(src_type, tgt_type)])

graph = reduce_reaction_nodes(graph)

plot_directed_graph(graph)
plt.show()