Executable Network Models of Integrated Multiomics Data

Abstract

Multiomics profiling provides a holistic picture of a condition being examined and captures the complexity of signaling events, beginning from the original cause (environmental or genetic), to downstream functional changes at multiple molecular layers. Pathway enrichment analysis has been used with multiomics data sets to characterize signaling mechanisms. However, technical and biological variability between these layered data limit an integrative computational analyses. We present a Boolean network-based method, multiomics Boolean Omics Network Invariant-Time Analysis (mBONITA), to integrate omics data sets that quantify multiple molecular layers. mBONITA utilizes prior knowledge networks to perform topology-based pathway analysis. In addition, mBONITA identifies genes that are consistently modulated across molecular measurements by combining observed fold-changes and variance, with a measure of node (i.e., gene or protein) influence over signaling, and a measure of the strength of evidence for that gene across data sets. We used mBONITA to integrate multiomics data sets from RAMOS B cells treated with the immunosuppressant drug cyclosporine A under varying O₂ tensions to identify pathways involved in hypoxia-mediated chemotaxis. We compare mBONITA's performance with 6 other pathway analysis methods designed for multiomics data and show that mBONITA identifies a set of pathways with evidence of modulation across all omics layers. mBONITA is freely available at https://github.com/Thakar-Lab/mBONITA.

Keywords: B cells; Boolean networks; chemotaxis; cyclosporine; hypoxia; multiomics; pathway analysis; proteomics.

Figure 1

mBONITA integrates information from multiple omics data sets to learn a consensus set of logic rules for prior knowledge networks (PKNs), simulate and perturb PKNs in silico, calculate condition-specific node modulation scores, and perform topology-based pathway analysis.

Figure 2

RAMOS B cells treated with cyclosporine A (CyA) and grown at different O₂ tensions were profiled at three molecular layers. (A) 1926 genes were profiled in all three omics data sets (median expression >0). (B) Correlations across different omics data sets. Expression data were processed and log₂-normalized as described in the Materials and Methods. Only genes profiled in all data sets were compared. Distinct experimental conditions are indicated by colors as shown in the figure annotations.

Figure 3

mBONITA identifies mechanisms of hypoxia-mediated chemotaxis from a multiomics data set from RAMOS B cells grown under three experimental conditions. All experiments used pathways downloaded from KEGG. (A) Inferred rule set sizes (ERS) for each omics data set and for the integrated analysis. Only nodes with in-degree ≥3 are shown. The mean (μ) and median (Mdn) of the ERS are shown for each data set. The red dashed line indicates μ. (B) Comparison of the mBONITA node importance scores learned from each experiment. Corr indicates the Pearson correlation coefficient. *** indicates that p < 0.01. (C) Number of differentially regulated (Benjamini–Hochberg corrected p < 0.05) KEGG pathways identified by mBONITA in the three contrasts.

Figure 4

Pathway-based prioritization of genes in a LSP1/HIF1A-centric signaling network with mBONITA. (A) Correlation between calculated node modulation scores N_m and its individual components (eq 3). ρ indicates the Pearson correlation coefficient (p < 0.01 in all cases). log₂FC = log₂ fold-change and SD = standard deviation. (B) The 50 nodes with highest variation in N_m across the three contrasts. Values above 2000 are grouped and indicated as >2000 on the color bar.

Figure 5

Comparison of mBONITA performance with the other pathway analysis methods. (A–C) Numbers of differentially regulated KEGG pathways identified from combination multiomics data in three contrasts: (A) 19% O₂, CyA– vs 1% O₂, CyA–. (B) 1% O₂, CyA+ vs 1% O₂, CyA–. (C) 19% O₂, CyA– vs 1% O₂, CyA+. (D) Significance of pathways known to be involved in the hypoxia-mediated response to CyA, for all three contrasts. Only pathways identified as significant from a combined data set by at least one method are shown. Pathways are defined as significantly modulated if the Benjamini–Hochberg corrected p < 0.05.