A Bayesian Approach for Identifying Multiple Regulatory Motifs Using Reversible Jump Markov Chain Monte Carlo

• Activity Number: 198
• Type: Contributed
• Date/Time: Tuesday, August 10, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Section on Bayesian Statistical Science
• Author(s): Sierra M. Li*+ and Jon C. Wakefield and Steve G. Self
• Companies: University of Washington and University of Washington and Fred Hutchinson Cancer Research Center
• Keywords: Bayesian inference ; transcription factor binding site (motif) ; Dirichlet distribution ; product multinomial ; Metropolis algorithm ; reversible jump Markov chain Monte Carlo

Abstract

Identification of transcription factor binding sites (regulatory motifs) plays an important role in gene discovery and characterization. Many methods have been proposed for motif detection, but none provide the flexibility to find repeated instances of multiple motif patterns of different widths. We have developed a unified Bayesian approach to simultaneous identification of multiple regulatory patterns based on a motif-specific probabilistic model. Unlike common approaches, our method does not require each motif pattern to occur in every sequence. The number of unique motifs is fixed and then the width, base composition, and occurrences (number and location) are evaluated for each motif within given ranges. We evaluate motif starting positions and their position-specific nucleotide distributions conditional on the number of motif pattern, motif widths and motif occurrences. Reversible jump Markov chain is used to handle the dimension change in the complex hierarchical parameterization. We show simulation results with sensitivity and specificity assessment. Finally, we use this method to identify regulatory motifs associated with genes involved in cell cycle regulation of yeast.