Jesper Lundström^1,2,3, Johan Björkegren^1,3 and Jesper Tegnér^2,3

¹The Computational Medicine group, Center for Molecular Medicine, Department of Medicine, Karolinska Institutet, Karolinska University Hospital, Solna, SE-171 76 Stockholm, Sweden. ²Division of Computa- tional Biology, Department of Physics, Linköpings Institute of Technology, Linköping University, SE-581 83 Linköping, Sweden. ³Clinical Gene Networks, Fogdevreten 2b, SE-17177, Stockholm, Sweden.

Abstract

Uncovering interactions between genes, gene networks, is important to increase our understanding of intrinsic cellular processes and responses to external stimuli such as drugs. Gene networks can be computationally inferred from repeated measurements of gene expression, using algorithms, which assume that each gene is controlled by only a small number of other proteins. Here, by extending the transcription network with cofactors (defined from protein-protein binding data) as active regulators, we identified the effective gene network, providing evidence of in-hubs in the gene regulatory networks of yeast. Then, using the notion that in-hub genes will be differentially expressed over several experimental conditions, we designed an algorithm, the HubDetector, enabling identification of in-hubs directly from gene expression data. Applying the HubDetector to 488 genome-wide expression profiles from two independent datasets, we identified putative in-hubs overlapping significantly with in-hubs in the effective gene network.