Large-scale inference of the transcriptional regulation of Bacillus subtilis

This paper addresses the inference of the transcriptional regulatory network of Bacillus subtilis. Two inference approaches, a linear, additive model and a non-linear power-law model, are used to analyze the expression of 747 genes from B. subtilis obtained using Affymetrix GeneChip® arrays under three different experimental conditions. A robustness analysis is introduced for identifying confidence levels for all inferred regulatory connections. Both the linear and non-linear methods produce candidate networks that share a scale-free or a “hub-and-spoke” topology with a small number of global regulator genes influencing the expression of a large number of target genes. The two computational approaches in tandem are able to identify known global regulators with a high level of confidence. The linear model is able to identify the interactions of highly expressed genes, particularly those involved in genetic information processing, energy metabolism and signal transduction. Conversely, the non-linear power-law approach tends to capture development regulation and specific carbon and nitrogen regulatory interactions.