Approximating High-Dimensional Simulations in Low-Dimensional Space, with Application to Microarray Prediction Error Estimation

• Activity Number: 673
• Type: Contributed
• Date/Time: Thursday, August 5, 2010 : 10:30 AM to 12:20 PM
• Sponsor: ENAR
• Author(s): Kevin Dobbin*+
• Companies: The University of Georgia
• Keywords: microarrays ; prediction error

Abstract

In high dimensional settings, often Monte Carlo and resampling procedures are associated with very high computational costs. This discourages use of the procedures in practice, and prohibits thorough evaluation of the properties of the procedures by simulation. This paper proposes a method for approximating these high dimensional procedures in a lower dimensional space. This method utilizes an approach common in evolutionary biology and MCMC to model the feature selection process. Modifications of the model for specific resampling settings are explored. An application using the method to reduce the computational cost of constructing confidence bounds on error rates of high dimensional microarray classifiers is shown.