Principal direction devising partitioning initialization of K-means clustering for discriminating among ovarian cancers while identifying the most salient genes involved

Abstract

This paper aims to cluster ovarian cancer patients described by gene expression data in order to discover the most discriminating genes responsible for the clustering. A combined use of Principal Direction Divisive Partitioning (PDDP) and bisecting K-means algorithms is applied to the investigated ovarian cancer dataset. The cascading of PDDP and bisecting K-means does successfully cluster ovarian cancer subjects and efficiently discovers salient genes needed to discriminate such clusters. The combined approach worked well on the automatic clustering of ovarian cancer patients depending merely on the gene expression information, and it has great potential for solving similar problems, like classifying leukemias or pancreatic tumors. The saliently identified genes may thus enhance relevant information for discriminating among ovarian cancers. In conclusion, the approach is shown to be a powerful one even in a complex multifactorial case like the intricate ovarian cancer discrimination.