A Generalized Filter Statistic for Integrated, Whole-Genome Analyses of Multiple Platforms to Characterize Several Groups of Similar Phenotype

• Activity Number: 340
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 10:30 AM to 12:20 PM
• Sponsor: Section on Nonparametric Statistics
• Author(s): Jeffrey Switchenko*+ and Jeanne Kowalski
• Companies: Emory University and Emory University
• Keywords: U-statistics ; Genomics

Abstract

With the increased availability of public domain data, there is a desire to examine several molecular features, such as expression, methylation and copy number, simultaneously, that characterize a specific patient group. Gene sets defined by several features could more rapidly provide novel insights underlying biologic mechanisms that define a patient group. Current attempts to integrate genome-wide data often include comparing results from several independent analyses of a single feature at a time or assume the existence of a relationship among features. We propose a unified, statistical inference framework for defining a genome-wide molecular signature based on several data types that is distinct to a specific group of patients. Our framework is based upon a composite, non-parametric filter statistic that has shown promise as a descriptive measure in previous applications. We formulate hypotheses that are tested based on this statistic by examining both its asymptotic properties using U-statistics and propose exact tests of significance based on a permutation approach. We illustrate our approach using multiple myeloma cell line data and breast cancer patient data.