|
Activity Number:
|
125
|
|
Type:
|
Contributed
|
|
Date/Time:
|
Monday, August 4, 2003 : 10:30 AM to 12:20 PM
|
|
Sponsor:
|
Section on Bayesian Stat. Sciences
|
| Abstract - #300433 |
|
Title:
|
Improving Bayesian Methods for Motif Discovery
|
|
Author(s):
|
Shane T. Jensen*+ and Jun S. Liu
|
|
Companies:
|
Harvard University and Harvard University
|
|
Address:
|
Department of Statistics, Cambridge, MA, 02138-2901,
|
|
Keywords:
|
metropolis optimization ; scoring functions ; motif-finding ; Bayesian Model ; missing data
|
|
Abstract:
|
An important application of statistical methods to genomics has been the problem of finding short sections (motifs) that are conserved in multiple DNA sequences. Motifs are of scientific interest as potential sites where proteins can bind directly to the DNA double-helix and regulate cell activity. The search for motif sites is nontrivial, since neither their locations nor the characteristics of the motif are known a priori. This problem is addressed by a Bayesian missing data model. This model is briefly reviewed along with stochastic implementation strategies using the Gibbs sampler. Stochastic algorithms can move freely in the parameter space to find a good motif signal, but the resulting motif will not necessarily be a posterior mode. We propose an algorithm called BioOptimizer that optimizes an appropriate scoring function to reduce noise in the motif signal found by stochastic algorithms. In addition, this scoring function optimization allows us to relax the usual assumptions of known motif width and motif abundance. The performance of BioOptimizer is evaluated by simulation studies and application to transcription factors in bacteria.
|
