An ensemble method of discovering sample classes using gene expression profiling

Cluster methods have been successfully applied in gene expression data analysis to address tumor classification. Central to cluster analysis is the notion of dissimilarity between the individual samples. In clustering microarray data, dissimilarity measures are often subjective and predefined prior to the use of clustering techniques. In this chapter, we present an ensemble method to define the dissimilarity measure through combining assignments of observations from a sequence of data partitions produced by multiple clusterings. This dissimilarity measure is then subjective and data dependent. We present our algorithm of hierarchical clustering based on this dissimilarity. Experiments on gene expression data are used to illustrate the application of the ensemble method to discovering sample classes.