A Hierarchical Mixture Model for State-Space Inference and Clustering

• Activity Number: 236
• Type: Contributed
• Date/Time: Monday, August 4, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Chandler Zuo*+ and Sunduz Keles
• Companies: University of Wisconsin-Madison and University of Wisconsin-Madison
• Keywords: clustering analysis ; mixture models ; E-M algorithm ; genetic regulatory network ; integrative analysis

Abstract

Modern epigenetic studies frequently require the integrative analysis of multiple datasets to discover association structures among observation units through their activation states across different experimental conditions. Such a problem is shared in many applications, such as inferring gene regulatory networks and annotating evolutionary modules. We developed the MBASIC (Matrix Based Analysis for State-space Inference and Clustering) framework to solve this general problem. MBASIC consists of two parts: state-space mapping, where the observations are mapped to a finite state-space by experiment-specific mixture models; and state-space clustering, where the units are clustered based on their inferred state-space profiles. Parameters of MBASIC can be estimated by an efficient EM algorithm. MBASIC flexibly adapts to different parametric distributions as well as heterogeneity in repeated experiments. We applied MBASIC to an integrative analysis of 170 ENCODE ChIP-seq datasets to discover clustered gene promoter regions. Compared to an naive approach using ENCODE peak annotation, MBASIC preserves more information in the raw data and discovers more significant clusters.