Reconstruction of regulatory network models and reconciliation with gene expression data

The reconstruction of genome-scale metabolic models from genome annotations has become a routine practice in Systems Biology research. The potential of metabolic models for predictive biology is widely accepted by the scientific community, but these same models still lack the capability to account for the effect of gene regulation on metabolic activity. Our focus organism, Bacillus subtilisis most commonly found in soil, being subject to a wide variety of external environmental conditions. This reinforces the importance of the regulatory mechanisms that allow the bacteria to survive and adapt to such conditions. We introduce a manually constructed and curated regulatory network model that describes the current state of knowledge of transcriptional regulation of B. subtilis. The model corresponds to an updated and enlarged version of the regulatory model of central metabolism originally proposed in 2008 [1]. We extended the original network to the whole genome by integration of information from DBTBS [2], a compendium of regulatory data that includes promoters, transcription factors (TFs), binding sites, motifs and regulated operons. Additionally, we consolidated our network with all the information on regulation included in the SporeWeb [3] and Subtiwiki [4] community-curated resources on B. subtilis. Finally, we reconciled our network with data from RegPrecise [5], which recently released their own less comprehensive reconstruction of the regulatory network for B. subtilis [6]. Our model describes 275 regulators, representing 30 different mechanisms of regulation such as TFs, RNA switch, Riboswitch, small regulory RNAs. Overall, regulatory information is included in the model for approximately 2500 of the ~4200 genes in B. subtilis 168. In an effort to further expand our knowledge of B. subtilis regulation, we reconciled our model with expression data. For this process we propose the concept of Atomic Regulon, as a set of genes that share the same ON and OFF gene expression profile across multiple samples of experimental data. Atomic regulon inference uses gene context information from operon predictions and curated SEED subsystems [7], in addition to expression data to infer regulatory interactions. Our studies with Atomic Regulation inference for multiple organisms revealed high confidence in the inferred regulons when compared with the popular gene regulatory network inference algorithm CLR [8]. We show how atomic regulons for B. subtilisare able to capture many sets of genes corresponding to regulated operons in our manually curated network. Additionally, we demonstrate how atomic regulons can be used to help expand/validate the knowledge of the regulatory networks by looking at highly-correlated genes missing regulatory information in the ARs. During this process we were also able to infer novel stimuli for hypothetical genes by exploring the genome expression metadata relating to experimental conditions, gaining insights into novel biology