Cellular phenotypes are determined by the differential activity of networks linking co-regulated genes. Available methods for the reverse engineering of such networks from genome-wide expression profiles have been successful only for the analysis of lower eukaryotes with simple genomes. Using a new method, ARACNe (Algorithm for the Reconstruction of Accurate Cellular Networks), we have reconstructed a genome-wide regulatory network from microarray expression profiles of normal, tumor-related, and experimentally manipulated human B cells. The resulting network, which includes about 129,000 interactions and 7,000 genes, suggest a hierarchical, scale-free network, where a few highly interconnected genes (hubs) account for most of the interactions.

Among the 5% largest hubs, the c-MYC proto-oncogene appears to control a hierarchically-organized sub-network involving known transcriptional targets as well as several new targets that were biochemically validated. Eleven out of twelve c-MYC targets predicted by the algorithm were validated in vivo using Chromatin Immunoprecipitation assays, a remarkable success rate for an in silico method. 

ARACNE was shown to outperform both Bayesian Networks and Relevance Networks based on extensive synthetic benchmarks, which were proposed specifically to compare reverse engineering methods. As a result, the new approach could be generally useful for the analysis of normal and pathologic networks in mammalian cells.