Building communities

micom will construct communities from a specification via a Pandas DataFrame. Here, the DataFrame needs at least two columns: “id” and “file” which specify the ID of the organism/tissue and a file containing the actual individual model.

To make more sense of that we can look at a small example. micom comes with a function that generates a simple example community specification consisting of several copies of a small E. coli model containing only the central carbon metabolism.

from micom.data import test_taxonomy

taxonomy = test_taxonomy()
taxonomy

	id	genus	species	reactions	metabolites	file
0	Escherichia_coli_1	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...
1	Escherichia_coli_2	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...
2	Escherichia_coli_3	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...
3	Escherichia_coli_4	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...
4	Escherichia_coli_5	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...

As we see this specification contains the required fields and some more information. In fact the specification may contain any number of additional information which will be saved along with the community model. One special example is “abundance” which we will get to know soon :) The logic behind requiring the community information in this table format is that this table can be appended as supplement to your project or publication as is and describes your community composition without any doubt.

In order to convert the specification in a community model we will use the Community class from micom which derives from the cobrapy Model class.

from micom import Community

com = Community(taxonomy)
print("Build a community with a total of {} reactions.".format(len(com.reactions)))

100%|██████████| 5/5 [00:00<00:00,  5.96models/s]

Build a community with a total of 495 reactions.

This includes the correctly scaled exchange reactions with the internal medium and initializes the external imports to the maximum found in all models. The original taxonomy is maintained in the com.taxonomy attribute.

Note that micom can figure out how to read a variety of different file types based on the extension. It curently supports:

• .pickle for pickled models
• .xml or .gz for XML models
• .json for JSON models
• .mat for COBRAtoolbox models

com.taxonomy

	id	genus	species	reactions	metabolites	file	abundance
id
Escherichia_coli_1	Escherichia_coli_1	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...	0.2
Escherichia_coli_2	Escherichia_coli_2	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...	0.2
Escherichia_coli_3	Escherichia_coli_3	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...	0.2
Escherichia_coli_4	Escherichia_coli_4	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...	0.2
Escherichia_coli_5	Escherichia_coli_5	Escherichia	Eschericia coli	95	72	/home/cdiener/code/micom/micom/data/e_coli_cor...	0.2

As you can notice we have gained a new column called abundance. This column quantifies the relative quantity of each individual in the community. Since we did not specify this in the original taxonomy micom assumes that all individuals are present in the same quantity.

The presented community here is pretty simplistic. For microbial communities micom includes a larger taxonomy for 773 microbial species from the AGORA paper. Here the “file” column only contains the base names for the files but you can easily prepend any path as demonstrated in the following:

from micom.data import agora

tax = agora.copy()
tax.file = "models/" + tax.file  # assuming you have downloaded the AGORA models to the "models" folder
tax.head()

	organism	id	kingdom	phylum	class	order	family	genus	species	oxygen_status	...	draft_created	platform	kbase_genome_id	pubseed_id	ncbi_id	genome_size	genes	reactions	metabolites	file
0	Abiotrophia defectiva ATCC 49176	Abiotrophia_defectiva_ATCC_49176	Bacteria	Firmicutes	Bacilli	Lactobacillales	Aerococcaceae	Abiotrophia	Abiotrophia defectiva	Facultative anaerobe	...	07/01/14	ModelSEED	NaN	Abiotrophia defectiva ATCC 49176 (592010.4)	592010.0	2041839	598	1069	840	models/Abiotrophia_defectiva_ATCC_49176.xml
1	Acidaminococcus fermentans DSM 20731	Acidaminococcus_fermentans_DSM_20731	Bacteria	Firmicutes	Negativicutes	Acidaminococcales	Acidiaminococcaceae	Acidaminococcus	Acidaminococcus fermentans	Obligate anaerobe	...	04/17/16	Kbase	kb|g.2555	Acidaminococcus fermentans DSM 20731 (591001.3)	591001.0	2329769	646	1090	903	models/Acidaminococcus_fermentans_DSM_20731.xml
2	Acidaminococcus intestini RyC-MR95	Acidaminococcus_intestini_RyC_MR95	Bacteria	Firmicutes	Negativicutes	Selenomonadales	Acidaminococcaceae	Acidaminococcus	Acidaminococcus intestini	Obligate anaerobe	...	08/03/14	ModelSEED	NaN	Acidaminococcus intestini RyC-MR95 (568816.4)	568816.0	2487765	599	994	827	models/Acidaminococcus_intestini_RyC_MR95.xml
3	Acidaminococcus sp. D21	Acidaminococcus_sp_D21	Bacteria	Firmicutes	Negativicutes	Selenomonadales	Acidaminococcaceae	Acidaminococcus	unclassified Acidaminococcus	Obligate anaerobe	...	06/29/12	ModelSEED	NaN	Acidaminococcus sp. D21 (563191.3)	563191.0	2238973	598	851	768	models/Acidaminococcus_sp_D21.xml
4	Acinetobacter calcoaceticus PHEA-2	Acinetobacter_calcoaceticus_PHEA_2	Bacteria	Proteobacteria	Gammaproteobacteria	Pseudomonadales	Moraxellaceae	Acinetobacter	Acinetobacter calcoaceticus	Aerobe	...	04/18/16	Kbase	kb|g.3519	Acinetobacter calcoaceticus PHEA-2 (871585.3)	871585.0	3862530	1026	1561	1165	models/Acinetobacter_calcoaceticus_PHEA_2.xml

5 rows × 24 columns

Saving and loading communities

Contructing large community models can be slow which is due to performance limitations of the solvers. In essence, adding a single variable/constraint to a model qih 10 variables is much faster than adding it to a model with 10 million variables. Thus, we recommend you save the constructed community in a serialized format afterwards which will be much faster in loading repetitively.

%time com = Community(taxonomy)

100%|██████████| 5/5 [00:00<00:00,  6.46models/s]

CPU times: user 784 ms, sys: 4.18 ms, total: 788 ms
Wall time: 785 ms

%time com.to_pickle("community.pickle")

CPU times: user 15.7 ms, sys: 4.13 ms, total: 19.8 ms
Wall time: 19.4 ms

from micom import load_pickle
%time com = load_pickle("community.pickle")

CPU times: user 152 ms, sys: 3.92 ms, total: 156 ms
Wall time: 154 ms

As we can see loading the model from the pickle format is much faster than creating it de novo.