A Nonhomogeneous Hidden Markov Model for Analyzing ChIP-chip Tiling Array Data

• Activity Number: 581
• Type: Contributed
• Date/Time: Wednesday, August 4, 2010 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Nancy Naichao Wang*+ and Terence Paul Speed
• Companies: Exponent, Inc. and University of California, Berkeley
• Keywords: non-stationary HMM ; tiling arrays ; data integration ; different probe designs ; multiple DNA-binding proteins ; shared targets

Abstract

There are two challenges in the analysis of ChIP-chip data that have not been satisfactorily met by existing methods. The first is to integrate replicate data across tiling arrays with different designs. The second is to identify the common sites of multiple DNA-binding proteins in a single analysis. We propose a nonhomogeneous hidden Markov model (HMM) for addressing both of these challenges. A set of curated positive and negative genomic regions from a C. elegans dataset was used in an ROC curve analysis to compare our approach to alternatives. When applied to each experiment separately, our method is comparable to the three best existing methods. When replicate data from arrays with different designs are integrated by our method, an improvement is achieved. Our method can analyze multiple proteins jointly and outperforms the approach of cross-listing the individual targets.