DNA-protein binding and gene expression patterns

Abstract

Although many clustering methods have been applied to analyze gene expression data, genes in the same cluster may have neither common functions nor common regulation. As a result, computational approaches have been developed to identify motifs in the regulatory regions of a cluster of genes or of genes with similar gene expression levels that are responsible for DNA-protein binding and similar gene expression levels. However, these motifs are neither sufficient nor necessary for a transcription factor to bind to the promoter region of a gene with these motif patterns. More recently, molecular methods have been developed to directly measure DNA-protein binding at the genomic level. In this article, we first evaluate the predictive power of computational approaches to predict DNAprotein binding from a study involving nine transcription factors in the cell cycle. We then compare how much variation in gene expression levels can be explained either by the observed DNA-protein binding or by the binding predicted through computational approaches. We find that current computational approaches may be limited both in predicting DNA-protein binding as well as in predicting gene expression levels. We also observe indirectly that the correspondence between gene expression levels and protein levels may be rather poor, which suggests that there may be difficulty in modeling genetic networks purely through gene expression data. To better understand gene expression patterns, an integrated approach to incorporating different kinds of information should be developed.