Gapped consensus motif discovery: evaluation of a new algorithm based on local multiple alignments and a sampling strategy
Abstract
We check the efficiency and faisability of a novel method designed for the discovery of a priori unknown motifs described as gaps alternating with specific regions. Such motifs are searched for as consensi of non homologous biological sequences. The only specifications required concern the maximal gap length, the minimal frequency for specific characters and the minimal percentage (quorum) of sequences sharing the motif. Our method is based on a cooperation between a multiple alignment method for a quick detection of local similarities and a sampling strategy running candidate position specific scoring matrices to convergence. This rather original way implemented for converging to the solution proves efficient both on simulated data, gapped instances of the so-called challenge problem, promoter sites in Dicot plants and transcription factor binding sites in E.Coli. Our algorithm compares favorably with the MEME and STARS approaches in terms of accuracy.