A computational genomics approach to the identification of gene networks

To delineate the astronomical number of possible interactions of all genes in a genome is a task for which conventional experimental techniques are ill-suited. Sorely needed are rapid and inexpensive methods that identify candidates for interacting genes, candidates that can be further investigated by experiment. Such a method is introduced here for an important class of gene interactions, i.e., transcriptional regulation via transcription factors (TFs) that bind to specific enhancer or silencer sites. The method addresses the question: which of the genes in a genome are likely to be regulated by one or more TFs with known DNA binding specificity? It takes advantage of the fact that many TFs show cooperativity in transcriptional activation which manifests itself in closely spaced TF binding sites. Such 'clusters' of binding sites are very unlikely to occur by chance alone, as opposed to individual sites, which are often abundant in the genome. Here, statistical information about binding site clusters in the genome, is complemented by information about (i) known biochemical functions of the TF, (ii) the structure of its binding site, and (iii) function of the genes near the cluster, to identify genes likely to be regulated by a given transcription factor. Several applications are illustrated with the genome of Saccharomyces cerevisiae , and four different DNA binding activities, SBF, MBF, a sub-class of bHLH proteins and NBF. The technique may aid in the discovery of interactions between genes of known function, and the assignment of biological functions to putative open reading frames.