Clustering Gene Expression Data Based on P Values

• Activity Number: 415
• Type: Contributed
• Date/Time: Thursday, August 12, 2004 : 8:30 AM to 10:20 AM
• Sponsor: Biometrics Section
• Author(s): Jun Li*+ and Rebecka J. Jornsten and Regina Y. Liu
• Companies: Rutgers University and Rutgers University and Rutgers University
• Keywords: clustering ; measure of similarity ; pairwise p values ; combining p values

Abstract

Clustering is an important task in many statistical analyses, including the analysis of microarray gene expression data, machine learning,and information retrieval. We use the analysis of microarray gene expression data to motivate and develop a new test-based clustering methodology that can reflect the exact experimental setup under which gene expression data are collected. We group genes by testing the equality or similarity of the condition-mean vectors and condition-variances. We use the p value from this test as a measure of similarity between two (or two groups of) genes, in which a small p value indicates that the mean-vectors and/or variance-vectors differ significantly. We view this measure of similarity as less arbitrary than the existing choices such as Euclidean, correlation etc. We cluster genes using all pairwise p values, building clusters in a bottom-up or top-down manner. For validation we use Fisher's method for combining p values to assign final p values to candidate clusters. Only clusters that satisfy the standard testing criteria are retained. To illustrate our clustering methodology, we use simulated and publicly available gene expression datasets.