The Genom of Homo sapiens.pdf

ANALYSIS OF HUMAN PROMOTERS 223nacchio and Rubin 2001), because different promotersevolve at different rates, multiple species would beneeded for narrowing down to short TFBSs. Initial successin yeast (Cliften et al. 2003; Kellis et al. 2003) maynot directly translate to human; novel integrated approacheswould be required to find functional cis elementseven if the number of mammalian genomes weredoubled.Integration, Combinatorial Analysis, andNetwork ReconstructionGenomic data is noisy; the best weapon for combatingnoise is signal correlation analysis. Combinatorial interactionamong TFs introduces correlation among theirbinding sites. Recently, there have been new motif-findingalgorithms, such as CO-Bind (GuhaThakurta andStormo 2001), that are designed specifically for detectingcorrelated motifs. Integration of evolutionary conservationwith word-pair analysis can yield a better regressionto expression data (Chiang et al. 2003).Integrating ChIP-chip and expression data at the singlemotif level has recently been attempted (Conlon et al.2003). We have developed two methods for studying cooperativityby integrating ChIP-chip data and microarrayexpression data. For a given pair of TFs, A and B, the firstmethod compares expression patterns of the targets ofboth TFs to that of A or B alone. If the former is more coherent(correlated), it is more likely that the two TFs areinteracting in the transcription regulation of their commontargets (Banerjee and Zhang 2003). The secondmethod further integrates with promoter sequence analysisin order not only to infer the interacting TFs, but alsoto assign their corresponding binding sites by iterativelyand exhaustively searching for significant TF combinationsand motif combinations up to the triplet level (M.Kato et al., in prep.). After analyzing over one hundredTF ChIP-chip data (Lee et al. 2002), we were able to reconstructthe yeast cell cycle transcriptional regulationnetwork so that (1) it extends the previous chain of singleregulators to an expanded chain of regulatory modules;(2) modules at adjacent phases often share a commoncomponent that can bridge the continuity of the cycle; (3)there are modules at specific checkpoints (branchpoints)that allow cell entry or exit of the cycle according to externalsignals (Fig. 4). Experimental verification is necessaryto confirm any network predictions (Segal et al.2003).Figure 4. Reconstructed yeast cell cycle transcriptional regulation network. (Adapted from M. Kato et al., in prep.)