Finding subtle motifs by branching from sample strings

Abstract

Many motif finding algorithms apply local search techniques to a set of seeds. For example, GibbsDNA (Lawrence et al., 1993) applies Gibbs sampling to random seeds, and MEME (Bailey and Elkan, 1994) applies the EM algorithm to selected sample strings, i.e. substrings of the sample. In the case of subtle motifs, recent bench-marking efforts show that both random seeds and selected sample strings may never get close to the globally optimal motif. We propose a new approach which searches motif space by branching from sample strings, and implement this idea in both pattern-based and profile-based settings. Our PatternBranching and ProfileBranching algorithms achieve favorable results relative to other motif finding algorithms. Availability: http://www-cse.ucsd. edu/groups/bioinformatics/software.html. © Oxford University Press 2003; all rights reserved.