Reaction-based Metabolome-Microbiome Integration

We first load some example data from github. This is data used also used in the manuscript. We simply load a .csv file with metabolite measurements and a group annotation file, together with a pre-computed metabolite-reaction graph. For more details on how to generate such a graph and what its structure, looks like see Network Generation and Using custom networks. The metabolite data should contain the samples in rows.

import networkx as nx
import matplotlib.pyplot as plt

from pymantra.network import (
    compute_reaction_estimates, add_reaction_estimates,
    compute_multiomics_associations, add_microbiome_associations,
    MultiOmicsLocalSearch
)
from pymantra.datasets import example_multiomics_enrichment_data


# loading example data
metabolite_data, microbiome_data, sample_groups, graph = \
    example_multiomics_enrichment_data()

Once all data is loaded, we first compute the linear models and the difference in residuals between samples groups by calling compute_reaction_estimates and add them to the graph in the required way using add_reaction_estimates. For how to use custom reaction models, see Custom Reaction Model Estimation.

# compute and add reaction estimates
residuals = compute_reaction_estimates(graph, metabolite_data, sample_groups)
add_reaction_estimates(graph, sample_groups, residuals)

Next, the associations between the reaction values and the microbiome measurements are computed by compute_microbiome_associations and add them to the graph by add_microbiome_associations. The correlations and p-values can also be analyzed seperately.

# compute and add microbiome-reaction associations
corr_associations, pvals = compute_multiomics_associations(
    residuals, microbiome_data, sample_groups, comparison=("0", "1"))
add_microbiome_associations(graph, sample_groups, corr_associations)

Finally, we can initialize a MultiOmicsLocalSearch object and run the local search. Afterwards, the local search object contains a number of functions to report and plot the results.

# generate local search object and run
# NOTE: these are randomly chose parameters
mo_lso = MultiOmicsLocalSearch(graph, "organism", 10., 1e-4, 4, 10, 10, 2)
mo_lso.run_local_search(n_threads=1, min_comp_size=4)

# report the results
print(f"Local Search solution: {mo_lso.solution}")
# saving results
mo_lso.solution[0].to_json("example_multiomics_solution.json")

mo_lso.plot_score_progression()
plt.show()

mo_lso.plot_subnetwork(graph)
plt.show()

# option omitting metabolite nodes
mo_lso.plot_subnetwork(node_types=nx.get_node_attributes(graph, "node_type"))
plt.show()

Entire Example Code

import networkx as nx
import matplotlib.pyplot as plt

from pymantra.network import (
    compute_reaction_estimates, add_reaction_estimates,
    compute_multiomics_associations, add_microbiome_associations,
    MultiOmicsLocalSearch
)
from pymantra.datasets import example_multiomics_enrichment_data


# loading example data
metabolite_data, microbiome_data, sample_groups, graph = \
    example_multiomics_enrichment_data()


# compute and add reaction estimates
residuals = compute_reaction_estimates(graph, metabolite_data, sample_groups)
add_reaction_estimates(graph, sample_groups, residuals)

# compute and add microbiome-reaction associations
corr_associations, pvals = compute_multiomics_associations(
    residuals, microbiome_data, sample_groups, comparison=("0", "1"))
add_microbiome_associations(graph, sample_groups, corr_associations)

# generate local search object and run
# NOTE: these are randomly chose parameters
mo_lso = MultiOmicsLocalSearch(graph, "organism", 10., 1e-4, 4, 10, 10, 2)
mo_lso.run_local_search(n_threads=1, min_comp_size=4)

# report the results
print(f"Local Search solution: {mo_lso.solution}")
# saving results
mo_lso.solution[0].to_json("example_multiomics_solution.json")

mo_lso.plot_score_progression()
plt.show()

mo_lso.plot_subnetwork(graph)
plt.show()

# option omitting metabolite nodes
mo_lso.plot_subnetwork(node_types=nx.get_node_attributes(graph, "node_type"))
plt.show()