An Approximate Empirical Bayes Model Selection Approach to Microarray Data Analysis

• Activity Number: 85
• Type: Invited
• Date/Time: Monday, August 4, 2008 : 8:30 AM to 10:20 AM
• Sponsor: ENAR
• Author(s): Harrison Zhou*+ and J.T. Gene Hwang and Dan Nettleton
• Companies: Yale University and Cornell University and Iowa State University
• Keywords: model selection ; microarray data analysis ; sparse inference

Abstract

Recent approach to model selection is to do variable selection and then estimate the coefficients of the selected variables. Some statistical procedures have been proposed to achieve these two goals simultaneously. In the regression context, it would be theoretically desirable if the proposed procedure dominates the ordinary least squared estimate, namely the natural procedure without variable selection. This property is called minimaxity. In this paper, we show that many well-known procedures fail to be minimax. We construct such minimax estimator which does variable selection as well. On the practical side, the estimators we construct perform as well as other well-known procedures even in the very sparse situations where the coefficients of variables are mostly zero. The procedure is easy to implement and computationally less intensive.