Skip to Main Content
Microarray experiments have been widely used to screen biological activities and cellular changes under different conditions at molecular level. Its ability to simultaneously monitor expression changes of thousands of genes has acquired its popularity but, at the same time, posed many challenging statistical and computational problems. Gene clustering problem is one of them. The purpose of gene clustering is to search for groups of genes with similar expression patterns, which likely have related biological functions or interactions. In this paper, popular gene clustering methods, including hierarchical clustering, AT-means, self-organizing maps, CLICK, model-based clustering and tight clustering, are introduced and compared with simulated data and a real data. The result shows the superior ability of tight clustering to deal with complex structure of microarray data. All the other methods encounter difficult problems of either scattered genes, local optimization or estimation of number of clusters. Tight clustering, on the other hand, provides an alternative to avoid these difficulties and systematically produce a sequence of tight clusters.