Comparative analysis of clustering methods for gene expresion data

Abstract

Large scale approaches, namely proteomics and transcriptomics, will play the most important role of the so-called post-genomics. These approaches allow experiments to measure the expression of thousands of genes from a cell in distinct time points. The analysis of this data can allow the the understanding of gene function and gene regulatory networks (Eisen et al., 1998). There has been a great deal of work on the computational analysis of gene expression time series, in which distinct data sets of gene expression, clustering techniques and proximity indices are used. However, the focus of most of these works are on biological results. Cluster validation has been applied in few works, but emphasis was given on the evaluation of the proposed validation methodologies (Azuaje, 2002; Lubovac et al., 2001; Yeung et al., 2001; Zhu & Zhang, 2000). As a result, there are few guidelines obtained by validity studies on which clustering methods or proximity indices are more suitable for the analysis of data from gene expression time series. Thus, this work performs a data driven comparative study of clustering methods and proximity indices used in the analysis of gene expression time series (or time courses). Five clustering methods encountered in the literature of gene expression analysis are compared: agglomerative hierarchical clustering, CLICK, dynamical clustering, k-means and self-organizing maps. In terms of proximity indices, versions of three indices are analysed: Euclidean distance, angular separation and Pearson correlation. In order to evaluate the methods, a k-fold cross-validation procedure adapted to unsupervised methods is applied. The accuracy of the results is assessed by the comparison of the partitions obtained in these experiments with gene annotation, such as protein function and series classification