Ultrahigh Dimensional Time Course Feature Selection

• Activity Number: 658
• Type: Invited
• Date/Time: Thursday, August 8, 2013 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Yi Li*+ and Peirong Xu and Lixing Zhu
• Companies: University of Michigan and East China Normal University and Hong Kong Baptist University
• Keywords: time course data ; longitudinal analysis ; generalized estimating equations ; variable selection ; sure screening property

Abstract

Statistical challenges arise from modern biomedical studies that produce time course genomic data with ultrahigh dimensions. For example, in a renal cancer study, the pharmacokinetic measures of a tumor suppressor (CCI-779) and expression levels of 12,625 genes were measured for each of 39 patients on 3 scheduled time points, with the goal of identifying predictive genes for pharmacokinetics over the time course. The resulting dataset defies analysis even with regularized regression. We propose a novel GEE-based screening procedure that only pertains to the specifications of the first two marginal moments and a working correlation structure. Instead of fitting separate marginal models as often adopted by the existing methods, our new procedure effectively reduces dimensionality of covariates by merely making a single evaluation of GEE functions. The new procedure is robust with respect to the mis-specification of correlation and enjoys theoretical readiness, which is further verified via intensive Monte Carlo simulations. We also apply the procedure to analyze the aforementioned renal cancer study.