Statistica Sinica 12(2002), 241-262

EVALUATION AND COMPARISON OF

CLUSTERING ALGORITHMS IN ANGLYZING ES

CELL GENE EXPRESSION DATA

Gengxin Chen$^1$, Saied A. Jaradat$^2$, Nila Banerjee$^1$,

Tetsuya S. Tanaka$^2$, Minoru S. H. Ko$^2$ and Michael Q. Zhang$^1$

$^1$Cold Spring Harbor Laboratory and

$^2$National Institutes of Health, U.S.A.

Abstract: Many clustering algorithms have been used to analyze microarray gene expression data. Given embryonic stem cell gene expression data, we applied several indices to evaluate the performance of clustering algorithms, including hierarchical clustering, $k$-means, PAM and SOM. The indices were homogeneity and separation scores, silhouette width, redundant score (based on redundant genes), and WADP (testing the robustness of clustering results after small perturbation). The results showed that the ES cell dataset posed a challenge for cluster analysis in that the clusters generated by different methods were only partially consistent. Using this data set, we were able to evaluate the advantages and weaknesses of algorithms with respect to both internal and external quality measures. This study may provide a guideline on how to select suitable clustering algorithms and it may help raise issues in the extraction of meaningful biological information from microarray expression data.

Key words and phrases: Cluster analysis, gene expression, microarray, mouse embryonic stem cell.