|
Abstract:
|
De Novo identification of transcription factor binding sites (TFBSs) is one of the most important problems in computational biology, as it provides valuable information towards elucidating gene regulatory networks. We focus on the search for repetitive but not identical patterns, called motifs, overpresented in the sequence data. A statistical approach requires a model for motif compositions, sites locations, motif widths, the number of motifs and the number of motif sites. This article proposes a novel Bayesian hierarchical model to incorporate all unknown features in motif discovery; inference over all unknown factors is associated with dimensional changes of the parameter space. Reversible jump Markov chain Monte Carlo allows parameter spaces with different dimensions to be explored. We apply out model to a set of upstream sequences of 20 co-expressed Saccharomyces cerevisiae genes.
|
