MotifCombinator: A Web Tool to Search for Combinations of Cis-Regulatory Motifs

Gene expression is controlled by combinatorial regulation of transcription factors and cis-regulatory elements in eukaryotes. Experimental procedures can determine several binding sites for selected transcription factors, but they are too laborious to be applied to large-scale studies. Computational methods are thus required to detect the combinatorial regulation at the genomic level. To reveal the combinatorial regulation, recent studies have developed computational methods to detect significant combinations of patterns (motifs) of cis-regulatory elements, using datasets of upstream sequences and expression levels from DNA microarrays, or binding information from ChIP-on-chip arrays [4]. One widespread type of such computational methods is to find combinations of motifs that specifically appear in upstream sequences of co-regulated genes, which are determined to be expressed by a certain threshold of expression levels. Another type of computational method is based on regression analysis between expression levels and motif scores (occurrence frequencies or weight matrix scores) in input sequences [2,3,6]. This type measures the goodness-of-fit of the regression for candidate motifs, and selects motifs with the best fitting scores. Well-known regression methods for this objective are the (multivariate) linear regression method [2,6] and the multivariate adaptive regression spline (MARS) method [3]. This type of computational method can take full advantage of information about expression levels because the methods do not compulsively dichotomize expression levels into whether genes are expressed or not. However, there is no web-based tool to systematically search motif combinations based on these regression methods. Moreover, these methods were developed mainly for simple eukaryotes like Saccharomyces cerevisiae; therefore, there are some limitations when they are applied to higher organisms with complex regulatory systems such as mammals. For example, they cannot practically handle combinations composed of more than two or three motifs. We implemented MotifCombinator, a web-based tool that can systematically search combinations of regulatory motifs based on regression methods. This tool is equipped with the two types of regression methods (the multivariate linear regression and MARS), and moreover it includes the logistic regression. It also employs the genetic algorithm to search combinations composed of more than two or three, or an arbitrary number of motifs. MotifCombinator will serve users to find combinations of regulatory motifs in organisms with complex regulatory systems.