Prediction of Motifs Based on a Repeated-Measures Model for Integrating Cross-Species Sequence and Expression Data

De novo identification of transcription factor binding sites (TFBS) is a challenging computational problem because TFBSs are relatively short sequences buried in long genomic regions. Earlier methods incorporated genome-wide expression data and promoter sequences into a linear-model framework, regressing expression on counts of putative TFBSs in promoters for a single species. More recently, it has been shown that examining sequence data across multiple species improves the prediction of TFBSs. In this work, we describe an extension of the single-species, linear-model framework for the analysis of paired cross-species sequence and expression data. A repeated measures model for gene-expression measurements across species is used, accounting for phylogenetic relationships among species through the error covariance structure. This multiple-species algorithm is applied to a data set of four yeast species grown under heat-shock conditions and comparisons are made to the single species algorithm. Using evaluations based on transcription factor binding strength and an independent source of expression data, we find the multiple species results show an improvement in the prediction of TFBS.