MICOMWeb: a website for microbial community metabolic modeling of the human gut

Abstract

MICOMWeb is a user-friendly website for modeling microbial community metabolism in the human gut. This website tackles three constraints when generating in silico metagenome-scale metabolic models: i) the prior Python user knowledge for metabolic modeling using flux balance analysis with the MICOM Python package, ii) predefined and user-defined diets to generate ad hoc metabolic models, and iii) the high-throughput computational infrastructure required to obtain the simulated growth and metabolic exchange fluxes, using real abundance from metagenomic shotgun or 16S amplicon sequencing; we present MICOMWeb's features to easily run in silico experiments as a functional hypothesis generator for experimental validation on three previously published databases. MICOMWeb has a constant run-time independent of the number of samples provided and database complexity. In practical terms, this behavior is upper-bounded by the sample with the greatest microbiota diversity, i.e., the sample with the largest metabolic reconstruction model size. The evidence suggests that the bigger the database, the better the MICOMWeb performs compared to MICOM in terms of consumed RAM (from 3.52 up to 7.13 folds) and total execution time (from 10.87 up to 205.05 folds).

Keywords: Systems biology; computational biology; metabolomics; microbiota; software computing.

Figure 1.

MICOMWeb overview. A) high-level front-to-backend infrastructure and SLURM processing strategy. B) Borromeo is the SLURM cluster used for MICOMWeb backend processing. A high-level overview, node, and storage specifications are presented.

Figure 2.

MICOMWeb to MICOM comparison. A) microbial community modelling with the MICOM Python package. B) MICOMWeb vs. MICOM reproducibility comparison on abundances, growth rates, number of metabolites, and reactions. D) The speed-up curve for the three example datasets using MICOM and MICOMWeb. An embedded zoom of the first 20 samples is also included.

Figure 3.

Comparative metabolic‐exchange profiling of diets. A) Principal component analysis (PCA) scatterplot of export fluxes for the top 50 metabolites, shown separately for type 2 diabetes and COVID−19 cohorts. The three dietary interventions (high fibre, high fat/protein, and fermented foods) are colour-coded. Rows correspond to cohort (top: COVID−19; bottom: Diabetes), and columns to flux direction (left: exports; right: imports). Component axes are labelled with the percentage of explained variance in the data. B) The heat map of the top 50 export fluxes for the three diets, where the cohort is presented at the top and bottom for COVID−19 and type 2 diabetes, respectively.

Figure 4.

Taxon‐specific growth‐rate distributions. Dot‐plot of per‐sample growth rates (plotted as 10^–growth_rate on a log scale) for the top 20 taxa by mean growth rate in each dietary intervention. Panels are arranged with rows for cohort (top: COVID−19–; bottom: Type 2 Diabetes– and columns for diet (high-fibre, high-fat/protein, fermented foods). Y-axis lists taxa in descending order of mean growth rate; individual points are jittered horizontally and coloured by taxon.