Clustering and Visualization of High-dimensional Biological Datasets Using a Fast Hma Approximation

In this paper, we reintroduce Hierarchical Mode Analysis(HMA), which was first proposed in 1968, as a powerful clustering algorithm for bioinformatics. The ability of HMA to find a compact hierarchy of a small number of dense clusters is very important in many bioinformatics problems (for example, when clustering genes in a set of gene-expression microarrays, where only a small number of genes related to the experimental context cluster well, while the rest need to be pruned). We also present two major improvements on HMA: a faster approximation algorithm, and a novel 2-D visualization scheme for high-dimensional datasets. These two improvements make HMA a powerful and promising new tool for many large, high-dimensional clustering problems in bioinformatics. We present empirical results on the Gasch dataset showing the effectiveness of our framework.