Integrative inference of transcriptional regulatory networks in Drosophila

Transcriptional regulatory networks describe regulatory relationships between transcription factors and target genes and constitute the core information processing machinery of a cell. Efforts of the modENCODE consortium have generated massive amounts of ChIP binding, expression, and chromatin datasets that capture different parts of the regulation machinery of the fly, Drosophila melanogaster. In this paper, we present both supervised and unsupervised methods of regulatory network inference that integrate these datasets to infer a functional regulatory network. Co-regulated genes in our networks are involved in similar functions, co-localize in tissues and interact in protein-interaction networks, much more than a sequencespecific motif or ChIP based network, suggesting that integrating different datasets is critical to inferring functional regulatory networks. The structures of our inferred networks exhibit a power-law degree distribution of target genes and are enriched in network motifs including feed-forward and feedback loops, which is consistent with hierarchical and modular organization of regulatory networks. Finally, the network structures can be used to predict expression and biological processes of target genes further validating that the connections between regulators and targets in our inferred networks are indeed functional and also demonstrating the predictive power of these networks.