Discussing a new Divisive Hierarchical Clustering algorithm

We present DHClus, a new Divisive Hierarchical Clustering algorithm developed to detect clusters with arbitrary shapes. Our algorithm is able to solve clustering problems defined by different scales, i.e. clusters with arbitrarily dissimilar densities, connectivity or between cluster distances. The algorithm not only works under this difficult conditions but it is also able to find the number of clusters automatically. This paper describes this new algorithm and then present results on real gene expression data. We compare the results of DHClus with other algorithms to provide a reference frame.