Identifying conserved discriminative motifs

Abstract

The identification of regulatory motifs underlying gene expression is a challenging problem, particularly in eukaryotes. An algorithm to identify statistically significant discriminative motifs that distinguish between gene expression clusters is presented. The predictive power of the identified motifs is assessed with a supervised Naïve Bayes classifier. An information-theoretic feature selection criterion helps find the most informative motifs. Results on benchmark and real data demonstrate that our algorithm accurately identifies discriminative motifs. We show that the integration of comparative genomics information into the motif finding process significantly improves the discovery of discriminative motifs and overall classification accuracy. © 2008 Springer Berlin Heidelberg.