Description:

To Define the Indefinable: Classification with Ultrahigh Dimensional Molecular Features

In ultrahigh-dimensional classification, using all features leads to poor performance. It is important to select a subset of important features to address the impact of dimensionality on classification. Most current ultrahigh-dimensional screening methods assume an independence rule, which is too restrictive for analyzing "-omic" data. In this talk I introduce a novel and computationally efficient multivariate screening and classification method for ultrahigh-dimensional data. Leveraging inter-feature correlations, the proposed method enables detection of marginally weak and sparse signals and recovery of the true informative feature set, and achieves optimal misclassification rates asymptotically. We show that the proposed procedure provides powerful discovery and classification boundaries. The performance of the proposed procedure is evaluated using simulation studies and an application of classifying post-transplantation rejection types based on genome-wide microarray profiles.