Fig. 1

Illustration of a genome-wide identification of modulator genes. In the present study we proposed a workflow to systematically identify key modulators from gene expression profiles. Briefly, for each putative modulator gene m, samples are sorted by its expression levels and the top/bottom 25% are defined as m-on/off samples. We designed four parameters to measure the modularity of m. The ACI score (parameter 1) measures the average change in normalized correlation coefficients between genome-wide gene interaction networks constructed in m-on and m-off samples. Focusing on the core subset of gene interactions, a m-modulated interaction network is built of significantly differentially correlated gene pairs called by MAGIC between the conditions. Three parameters, namely number of nodes, number of edges, and connectivity (i.e., average node degree), are employed to measure the scale and information flow of the core network. The procedures are performed iteratively to analyze each gene in the expression dataset