Parametric validity index of clustering for microarray gene expression data

An important area of genomic signal processing is microarray gene expression data analysis, which employs clustering algorithms to group individual genes or samples in a population. Due to the non-unique nature of clustering, the cluster validation is necessary for evaluating the results of clustering algorithms. In this paper, we propose a parametric validity index (PVI) which employs two tunable parameter α and β to control the proportions of objects being taken into account to calculate the dissimilarities. There are two advantages of the proposed PVI: on one hand, its computational complexity is low, and on the other hand, it has flexibility of tuning the parameters to meet different datasets, especially the microarray datasets. The PVI can be averaged over a range of values of α and β. We investigate the new PVI for assessing five clustering algorithms in four microarray datasets. The experimental results appear to suggest that the proposed PVI has relatively robust performance and provides fairly accurate judgements.