A Method for Comparing Clustering Structures, with Application to DNA Microarrays

• Activity Number: 273
• Type: Invited
• Date/Time: Tuesday, August 5, 2003 : 2:00 PM to 3:50 PM
• Sponsor: Biometrics Section
• Author(s): Ilana Belitskaya*+
• Companies: Stanford University
• Keywords: DNA microarrays ; cluster analysis ; clustering structure ; interpoint-distance-based graph

Abstract

Various methods have been proposed in the literature for comparing partitions or dendrograms constructed on n objects. Such methods include measures of partitional agreement, such as the Rand statistic [Rand, 1971], and measures based on cophenetic proximity for comparing hierarchies, described in Jain and Dubes [1988]. These measures are used for a variety of purposes. One is to compare clustering structures of objects based on different subsets of the variables. However, the methods proposed in the literature depend on the particular clustering algorithm used. We propose a method based on interpoint-distance-based graphs for comparing clustering structures of objects that does not depend on clustering procedures.This method can be used to compare clustering structures of n objects based on different sets of variables or to assess association between a known partition of objects and the underlying clustering structure of objects based on a set of variables. The motivating application is to gene expression data.