Guiding motif discovery by iterative pattern refinement

Abstract

In this paper, we demonstrate that the performance of a motif discovery algorithm can be significantly improved by embedding it into a novel framework that effectively guides the motif discovery process. The framework is also general enough to allow any statistical motif discovery algorithm to be used. Motivation for this research comes from the fact that the statistical significance of patterns depends on the background probability which is largely determined by input sequences. Our framework guides motif discovery by inputting subsequences to an existing motif discovery algorithm, rather than using entire sequences. Subsequences are determined by motifs discovered using existing motif discovery and search algorithms. Then this technique is iteratively applied until convergence. A starting set of patterns is discovered by a simple, but effective pattern set generation algorithm. Our framework was implemented using MEME and MAST and tested with 108 PROSITE patterns. The result demonstrates that our framework significantly improves the performance of MEME.