Analyzing metabolic interactions

Whereas the default results will show you which taxon consumes and produces which metabolite it is not immediately apparent which metabolic interactions that implies. Thus, we provide some helpers to quantify and summarize metabolic interactions between taxa.

It should be noted that MICOM provides mechanistic interactions, thus they differ quite abit from correlations. First, they are calculated on a per sample basis and can thus differ between samples as well. They are also non-symmetric and directed and thus qualify as ecological interactions. The strategy used in grow workflow might also affect the predicted interactions. The most conservative (least interactions) will be predicted with parsimonious FBA because it will also minimize inter-taxon fluxes. The other strategies are somewhat more permissive for inter-taxon fluxes but may also include futile cycles (which will not appear in parsimonious FBA).

Finally all-versus-all interactions can be somewhat slow for larger data sets due to the inherent combinatorial explosion. The complexity will scale with \(n_{samples} \cdot n^2_{taxa} \cdot n_{metabolites}\).

Calculating focal interactions

Interactions are obatined from a GrowthResults object as obtained by the grow workflow. By default they are based on a single taxon of interest (called a focal taxon) for which we will calculate all metabolic interactions with all other taxa in all samples. MICOM stratifies interactions into 3 ecological types shown below.

MICOM interaction types

Let’s see what that looks like by using a larger example result.

[1]:

from micom.data import test_results
from micom.interaction import interactions

results = test_results()
ints = interactions(results, taxa="s__Akkermansia_muciniphila")
ints.sort_values(by="flux", ascending=False).head()

[1]:
metabolite focal partner class flux sample_id name molecular_weight C_number N_number ... kegg.compound lipidmaps metanetx.chemical pubchem.compound reactome sbo seed.compound chebi smiles reaction
2115 ala_L[e] s__Akkermansia_muciniphila s__Bacteroides_fragilis provided 11.238551 S_SRR5935812 L-alanine 89.09318 3 1 ... C00041 NaN MNXM1105732 5950.0 NaN SBO:0000247 cpd00035 CHEBI:16977 NaN EX_ala_L(e)
2141 lac_D[e] s__Akkermansia_muciniphila s__Bacteroides_fragilis provided 11.141526 S_SRR5935812 D-lactate 89.07000 3 0 ... C00256 NaN MNXM731835 61503.0 NaN SBO:0000247 cpd00221 CHEBI:42111 NaN EX_lac_D(e)
1080 acald[e] s__Akkermansia_muciniphila s__Escherichia_coli received 9.277781 S_SRR5935769 Acetaldehyde 44.05256 2 0 ... C00084 NaN MNXM75 177.0 NaN SBO:0000247 cpd00071 CHEBI:15343 NaN EX_acald(e)
1120 pro_L[e] s__Akkermansia_muciniphila s__Escherichia_coli received 9.208372 S_SRR5935769 L-proline 115.13046 5 1 ... C00148;C000763 NaN MNXM114 145742.0 NaN SBO:0000247 cpd00129 CHEBI:17203 NaN EX_pro_L(e)
1094 etoh[e] s__Akkermansia_muciniphila s__Escherichia_coli received 8.980545 S_SRR5935769 Ethanol 46.06844 2 0 ... C00469 NaN MNXM734299 702.0 NaN SBO:0000247 cpd00363 CHEBI:16236 NaN EX_etoh(e)

5 rows × 24 columns

So you see the individual metabolite flux between the pair of taxa and the interaction class in each single sample. This is quite detailed and can tell you a lot, but we can also summarize it to the overall fluxes between two taxa in a sample.

[2]:

from micom.interaction import summarize_interactions

summary = summarize_interactions(ints)
summary.head()

[2]:
sample_id focal partner class flux mass_flux C_flux N_flux n_ints
0 S_SRR5935769 s__Akkermansia_muciniphila s__Alistipes_finegoldii co-consumed 7.349260 0.535659 16.389235 2.699664 8
1 S_SRR5935769 s__Akkermansia_muciniphila s__Alistipes_finegoldii provided 3.180756 0.342653 13.209727 2.560137 6
2 S_SRR5935769 s__Akkermansia_muciniphila s__Alistipes_finegoldii received 1.513632 0.321052 13.030612 2.967128 6
3 S_SRR5935769 s__Akkermansia_muciniphila s__Alistipes_onderdonkii co-consumed 2.329465 0.259526 8.734520 1.738114 8
4 S_SRR5935769 s__Akkermansia_muciniphila s__Alistipes_onderdonkii provided 0.906143 0.112754 5.490118 0.764795 4

This calculates the overall flux in each class for the pair of taxa. It also provides some more meaningful eestimates of flux such as the exchanged mass, carbon or nitrogen. So in this example you see that Akkermansia competes for most of the mass with Alistipes but actually receives more nitrogen from Alistipes than it competes for in that sample.

Calculating all interactions

It’s also possible to calculate all interactions, just be aware that this (1) generates a lot of data and (2) will take a while for larger data sets. You can parallelize the analysis over each taxon by providing the the threads argument. Simply provide None for the taxa to obtain all interactions.

[3]:

full = interactions(results, taxa=None, threads=8)
full.shape



















[3]:







(1115688, 24)





This generates quite a lot of results, but we can again use a summary.




[4]:





full_summary = summarize_interactions(full)
full_summary.shape









[4]:







(117108, 9)









Metabolic Exchange Score


For a slightly more global view on exchanges we also provide calculation of the Metabolic Exchange Score (MES) by Marcelino et al., which is decribed in detail here. The MES is the geometric mean of the number of producers P and consumers C of a metabolite m in a single sample i, given by:





\[MES^i_m = 2\cdot\frac{P^i_m\cdot C^i_m}{P^i_m + C^i_m}\]




This can be interpreted as a normalized number of all observed metabolic interactions (becuse \(P\cdot C\) is the number of all possible directed combinations of producers and consumers). So it is a measure of cross-feeding.


It can also be calculated very fast for large data sets, so let’s go.




[5]:





from micom.interaction import MES

scores = MES(results)
scores.head()









[5]:











  
    

      
      metabolite
      sample_id
      MES
      name
      molecular_weight
      C_number
      N_number
      bigg.metabolite
      biocyc
      hmdb
      ...
      kegg.compound
      lipidmaps
      metanetx.chemical
      pubchem.compound
      reactome
      sbo
      seed.compound
      chebi
      smiles
      reaction
    

  
  
    

      0
      12dhchol[e]
      S_SRR5935769
      4.800000
      12-Dehydrocholate
      405.54758
      24
      0
      12dhchol
      NaN
      NaN
      ...
      NaN
      NaN
      NaN
      NaN
      NaN
      SBO:0000247
      NaN
      NaN
      NaN
      EX_12dhchol(e)
    

    

      1
      12dhchol[e]
      S_SRR5935812
      4.800000
      12-Dehydrocholate
      405.54758
      24
      0
      12dhchol
      NaN
      NaN
      ...
      NaN
      NaN
      NaN
      NaN
      NaN
      SBO:0000247
      NaN
      NaN
      NaN
      EX_12dhchol(e)
    

    

      2
      12dhchol[e]
      S_SRR5935816
      3.428571
      12-Dehydrocholate
      405.54758
      24
      0
      12dhchol
      NaN
      NaN
      ...
      NaN
      NaN
      NaN
      NaN
      NaN
      SBO:0000247
      NaN
      NaN
      NaN
      EX_12dhchol(e)
    

    

      3
      12dhchol[e]
      S_SRR5935843
      4.444444
      12-Dehydrocholate
      405.54758
      24
      0
      12dhchol
      NaN
      NaN
      ...
      NaN
      NaN
      NaN
      NaN
      NaN
      SBO:0000247
      NaN
      NaN
      NaN
      EX_12dhchol(e)
    

    

      4
      12dhchol[e]
      S_SRR5935924
      4.444444
      12-Dehydrocholate
      405.54758
      24
      0
      12dhchol
      NaN
      NaN
      ...
      NaN
      NaN
      NaN
      NaN
      NaN
      SBO:0000247
      NaN
      NaN
      NaN
      EX_12dhchol(e)
    

  



5 rows × 21 columns






All of the computed interaction measures also have matching visualizations.