Chapter 4 - Hierarchical k-Means: A Hybrid Clustering Algorithm and Its Application to Study Gene Expression in Lung Adenocarcinoma

Abstract

DNA microarray technology is a very useful tool to get information for various biological processes by providing a convenient way to investigate expression levels of thousands of genes in a collection of related samples. Moreover, as a discovery tool, DNA microarrays have the potential to enhance the understanding of the molecular basis of different diseases by simultaneously providing information on thousands of genes. Researchers using DNA microarray technology often find it interesting and meaningful to group genes with similar expression patterns together. k-means and hierarchical clustering are two widely used clustering algorithms used for this purpose. However, both of these algorithms suffer from some limitations. Although the k-means clustering algorithm is an efficient algorithm in producing tight clusters, it requires a prespecified number of clusters as input, and the performance of this clustering algorithm mostly depend on this specification. Hierarchical clustering, on the other hand, produces a hierarchy of clusters that is very informative; however, this advantage comes at the cost of low efficiency. This chapter describes a new clustering algorithm called hierarchical k-means that was used to overcome these limitations. In this algorithm, the number of clusters is determined from the hierarchy of clusters produced by the hierarchical clustering algorithm, and then this information is fed into k-means to produce the final clusters. After applying this proposed algorithm to microarray data of lung adenocarcinoma followed by the functional investigation of representative genes from the group of normal tissues and KRAS mutation tissues, we found that this proposed algorithm can group genes with similar expression patterns together. Therefore, in future studies, this proposed algorithm can definitely be used for clustering microarray data.