Biclustering with the EM Algorithm

• Activity Number: 652
• Type: Contributed
• Date/Time: Thursday, August 2, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Prabhani Kuruppumullage Don*+ and Bruce G. Lindsay and Francesca Chiaromonte
• Companies: Penn State University and Penn State University and Penn State University
• Keywords: Bi-clustering ; Composite likelihood ; Mixture ; Gradient function

Abstract

Cluster analysis is the identification of natural groups in data. Most clustering applications focus on one-way clustering, grouping observations that are similar to each other based on a set of features, or features that are similar to each other across a given set of observations. With large amount of data arising from applications such as gene expression studies, text mining studies, etc., there has been renewed interest in bi-clustering methods that group observations and features simultaneously. In one-way clustering, it is known that mixture-based techniques using the EM algorithm provide better performance, as well as an assessment of uncertainty. In bi-clustering however, evaluating the mixture likelihood using an EM algorithm is computationally infeasible and approximations are essential. In this work, we propose an approach based on a composite likelihood approximation and a nested EM algorithm to maximize the likelihood. We study the convergence of the algorithm using simulations. Further, we propose to use the EM algorithm on a mixture gradient function to evaluate the suitability of the number of components.