Bicluster analysis of genome-wide gene expression

Abstract

A number of biclustering approaches have been developed to mitigate the limitations of standard clustering algorithms. They have different problem formulation, search strategy and computational complexity. We proposed a new biclustering method based on the framework of market basket analysis in which a bicluster is described as a frequent itemset. As a feasibility test, we compared it with several standard clustering algorithms on a genome-wide yeast microarray dataset, and it showed very promising results. We later did a comparison between our approach and various current biclustering methods, following a systematic evaluation procedure recently published. The experimental results demonstrate that our new method outperforms the others.