Fractal Clustering for Microarray Data Analysis

Abstract

DNA microarray experiments generate a substantial amount of information about global gene expression. Gene expression profiles can be represented as points in multi-dimensional space. It is essential to identify relevant groups of genes in biomedical research. Clustering is helpful in pattern recognition in gene expression profiles. Some clustering techniques have been introduced. However, these traditional methods mainly utilize shape-based assumption or distance metric to cluster the points in multi-dimension linear Euclidean space. Poor consistence with the functional annotation of genes is shown in their validation study. A fractal clustering method to cluster genes using intrinsic (fractal) dimension from modern geometry is proposed. Fractal dimension is used to characterize the degree of self similarity among the points in the clusters. The main idea of fractal clustering is to group points in a cluster in such a way that none of the points in the cluster changes the cluster's intrinsic dimension radically. Hausdorff fractal dimension is computed through the means of the box-counting plot algorithm, since it is the fastest and also robust enough. This method is assessed using validation assessment using public microarray dataset. It shows that this method is superior in identifying functional related gene groups than other traditional methods.