Bayesian analysis of array CGH data via Markov chain Monte Carlo simulation

Zhengdong Zhang; Molecular Biophysics and Biochemistry Dept., Yale University
Friday, 29th September 2006 (4pm)
Windsor Building, Room 0-04

Typical array comparative genomic hybridization (CGH) data consist of log-ratios of normalized intensities from the test sample to those from the reference sample, indexed by the probes' genomic locations. The goal of array-CGH data analysis is to detect copy number variations (CNVs) by identifying regions with log-ratios that are consistently higher or lower than the normalized base-line. Here we present a Bayesian statistical framework to analyze array-CGH data. Treating parameters that define the underlying genomic copy number variation encoded in the data as random variables, our Bayesian approach derived a posterior distribution of those parameters given the observed data. We discuss how our model was derived and implemented, and the empirical results from applying our method to both simulated and real array-CGH data.