Consistent Biclustering

• Activity Number: 176
• Type: Contributed
• Date/Time: Monday, July 30, 2012 : 10:30 AM to 12:20 PM
• Sponsor: Section on Statistical Learning and Data Mining
• Author(s): Cheryl J. Flynn*+ and Patrick O Perry
• Companies: New York University and NYU Stern
• Keywords: Biclustering ; Consistency ; Microarray Data ; Directed Networks ; Term-Document Matrices

Abstract

Biclustering, the process of simultaneously clustering observations and variables, is a popular and effective tool for finding structure in a high-dimensional dataset. A variety of biclustering algorithms exist, and they have been applied successfully to data sources ranging from gene expression arrays to term-document matrices. Currently, while biclustering appears to work well in practice, there have been no theoretical guarantees about its performance. We address this shortcoming with a theorem providing sufficient conditions for asymptotic consistency when both the number of observations and the number of variables in the dataset tend to infinity. This theorem applies to a broad range of data distributions, including Gaussian, Poisson, and Bernoulli. We demonstrate our results through a simulation study and with examples drawn from microarray analysis, document clustering, and network analysis.