Fig 1.
Characterization of iModulons derived from five independent gene expression datasets.
(A) Schematic illustration of the workflow applied to each data set. (B) Descriptions of the three classes of characterized iModulons. The first column contains histograms illustrating the distribution of gene weightings in each of three independent components (ICs) from the RNAseq-1 dataset. Genes outside of a threshold (in red) belong to an “iModulon”. The second column illustrates the biological interpretation of the iModulon types. iModulons are characterized by comparing their genes with known regulons, ontological annotations, and genotypes. The third column displays the ICA-computed activity levels for the selected ICs across all 278 conditions in the RNAseq-1 dataset. (C) Bar chart describing the four types of iModulons computed from each of the five gene expression datasets. (D) Comparison of three Functional iModulons, each derived from decomposition of a different dataset. Asterisks indicate genes annotated with the GO term “lipopolysaccharide core region biosynthetic process”.
Table 1.
Summary of transcriptomic datasets.
Fig 2.
Comparison of IC gene weightings for iModulons enriched with genes in the CysB regulon.
(A) Scatter plot between IC gene weightings for CysB-linked iModulons across all five datasets. Genes in the published CysB regulon are colored in red. For comparisons in the first column involving the MA-3 dataset, genes regulated by any of the following regulators are colored in blue: MetJ, TrpR, GlpR, ArgR, Lrp, CysB, leu-tRNA-mediated transcriptional attenuation, or thiamine riboswitch. All other genes are black. Dashed lines indicate iModulon thresholds. Gray solid lines indicate the 45-degree line of equal gene weightings. (B) Scatter plot between IC gene weightings for the CysB iModulon in MA-3 compared to a linear combination of 10 ICs from RNAseq-1 (see S1 Table). Color scheme is identical to panel (A).
Fig 3.
The iModulon structure is conserved across five datasets.
(A) Reciprocal best hit (RBH) graph indicating iModulons as nodes and RBHs as edges. Node color indicates the source dataset for the iModulon, as denoted for the CysB iModulon. Node shape indicates the iModulon category. Edge thickness and darkness indicate the gene weighting similarity. Clusters are labelled with the regulator(s) that are linked to the iModulons in the cluster, if available, and grouped by level of iModulon reproducibility. (B) Pie chart describing the categories of all iModulons shown in the RBH graph. (C) Pie chart describing the types of edges in the RBH graph. (D) Heatmap indicating how many iModulons from each dataset and category were in the RBH graph.
Fig 4.
iModulon activities capture the effects of mutations from adaptive evolution to multiple antibiotics.
(A) Bar chart of activities for the MarA/Rob iModulon. Individual points represent biological replicates. The heatmap below the bar chart shows the presence of mutations in the specified gene for the specified strain. Strain names are described in Lazar et al [34]. (B) Regulatory network for antibiotic resistance in E. coli. Black arrows indicate activation, and red arrows represent repression. Auto-regulation is not shown. (C) Bar chart of activities for the SoxS iModulon, similar to panel (A). Heatmap shows the presence of mutations in soxR. (D) Venn diagram showing the overlap of genes in the MarA/Rob iModulon and the SoxS iModulon. (E) Bar chart of activities for the RcsAB iModulon, similar to panel (A). Heatmap shows the presence of a genomic inversion. (F) Schematic illustration of the genomic inversion upstream of lon. The inversion decreases lon expression, resulting in longer residency times for MarA, SoxS, and RcsA.
Fig 5.
Decomposition of combined compendium results in increased resolution (A) Schematic illustration of data integration. (B) Pie chart showing the categories of the 181 iModulons from the combined dataset. (C) Heatmap illustrating which iModulons from the combined compendium were matched to iModulons in individual datasets in the RBH graph. The total number of matches for each iModulon is shown in blue. Category of each iModulon is shown below. (D) Heatmap illustrating how many iModulons from each category and dataset were not matched to an iModulon in the full compendium in the RBH graph. (E) Pie chart illustrating the fraction of expression difference resulting from ppGpp-RNAP binding explained by each iModulon. The norfloxacin-related iModulon is described in S7E Fig. (F) Boxplot of the Central Dogma iModulon activities for expression profiles in the RNAseq-1 dataset. (G) Principal component loadings of the combined compendium expression levels without batch correction. (H) Principal component loadings of the iModulon activities from the combined compendium.
Fig 6.
Predicting regulons using big data.
(A) Pie chart displaying the number of Regulatory, Functional and Uncharacterized iModulons extracted from the COLOMBOS E. coli compendium. (B) Venn diagram illustrating the number of iModulons shared between the COLOMBOS compendium and the combined dataset discussed in Fig 5. (C) Histogram of the overlap coefficients between the 131 shared iModulons between COLOMBOS and the combined dataset. (D) Scatter plot of the iModulon gene weights for the putative HprR iModulon. Purple genes are in both the iModulon from COLOMBOS and the iModulon from the combined dataset iModulon. Red genes are only in the iModulon from COLOMBOS, and blue genes are only in the iModulon from the combined dataset. The dashed lines indicate iModulon thresholds, and the gray diagonal line is the 45-degree line. (E) Schematic representation of the genes near hprR. (F) Bar chart of the putative HprR iModulon activities from GEO dataset GSE35371. (G) Scatter plot of the iModulon gene weights for an uncharacterized iModulon. Colors are identical to panel (d). (H and I) Relative iModulon activities of the iModulon from panel (G) from GEO datasets GSE21839 and GSE55365, respectively. Each dataset is centered to its own reference condition, so relative activities cannot be compared across bar charts. (J) Scatter plot of the iModulon gene weights for the antibiotic-responsive uncharacterized iModulon. (K and L and M) Bar chart of the antibiotic-responsive iModulon activities from GEO datasets GSE31140, GSE37026, and GSE10158.