The Spherical K-means algorithm

Clustering on the hypersphere with the Spherical K-means algorithm

Clustering is one of the most important data mining techniques used to extract useful information
from microarray data. Microarray data sets can be either clustered by samples or by genes. In this
research we focus on the gene clustering problem. The objective of gene clustering is to group genes with
similar expression patterns together with the common belief that those genes often have similar functions,
participate in a particular pathway or response to a common environmental stimulus. Although hundreds
of clustering algorithms exist, the very simple K-means and its variants remain among the most widely
used algorithms for gene clustering by biologists and practitioners. This surprising fact may be attributed
to its especial ease of implementation and use.
When microarray data are normalized to zero mean and unit norm, a variant of the K-means algorithm that works with the normalized data would be more suitable. Since the data points are on a unit hypersphere, the algorithm is called
the Spherical K-means algorithm (SPK-means).