Clustering Gene Expressions in the Presence of Scatter

• Activity Number: 340
• Type: Contributed
• Date/Time: Tuesday, July 31, 2007 : 2:00 PM to 3:50 PM
• Sponsor: Section on Statistical Computing
• Author(s): Ivan Ramler*+ and Ranjan Maitra
• Companies: Iowa State University and Iowa State University
• Keywords: Gene Expression ; k-mean-directions ; Bayes Information Criterion

Abstract

A new methodology is proposed for clustering gene expression datasets in the presence of scattered observations. These are defined to be observations that are unlike any other, so traditional approaches that force them into groups can lead to all-around erroneous conclusions. Our suggested approach is an iterative scheme which proceeds by building the core for each cluster around the centers, identifies points outside as scatter and updates the method until convergence. In the absence of scatter, the algorithm reduces to a k-means algorithm designed for constrained directional data. We also provide methodology to initialize the algorithm as well as to estimate the number of clusters in the dataset. Results on several sets of test experiments show excellent performance. The methodology is applied to gene expression data on the diurnal starch cycle of Arabidposis L. Heynth.