Hierarchical Clustering of Gene Expression Data with Divergence Measure

Hierarchical clustering is a commonly used and valuable approach in clustering analysis. However it depends on the measure used to assess similarity between samples. Two frequently adopted distance measures are Euclidean distance (L₂- norm) and city-block distance (L₁-norm), and they do not take into account special characteristics of data at hand. In this paper, considering the nonnegativity of gene expression data, we apply a generalized Kullback-Leibler (KL) divergence to measure the similarity in hierarchial clustering analysis. Experimental results on several real cancer related gene expression datasets show that the proposed KL divergence outperforms both L₂ and L₁ distances.