Short Segment Frequency Equalization: A Simple and Effective Alternative Treatment of Background Models in Motif Discovery

摘要

One of the most important pattern recognition problems in bioinfor-matics is the de novo motif discovery. In particular, there is a large room of improvement in motif discovery from eukaryotic genome, where the sequences have complicated background noise. The short segment frequency equalization (SSFE) is a novel treatment method to incorporate Markov background models into de novo motif discovery algorithms, namely Gibbs sampling. Despite its apparent simplicity, SSFE shows a large performance improvement over the current method (Q/P scheme) when tested on artificial DNA datasets with Markov background of human and mouse. Furthermore, SSFE shows a better performance than other methods including much more complicated and sophisticated method, Weeder 1.3, when tested with several biological datasets from human promoters.