The study of gene interactions is an important research area in biology and grouping genes with similar expression profiles to clusters is a first step towards a better understanding of their functional relationships. In Kaba et al. 2007, a new clustering approach was presented, using gene interaction graphs to model this data, and decomposing the graphs by means of clique minimal separators. A clique separator is a clique whose removal increases the number of connected components of the graph; the decomposition is obtained by repeatedly copying a clique separator into the components it defines, until only subgraphs with no clique separators are left: these subgraphs will be our clusters. The advantage of our approach is that this decomposition can be computed efficiently, is unique, and yields overlapping clusters. For that, the similarity between each pair of genes is estimated by a distance function, then a family of gene interaction graphs is constructed by choosing several thresholds, where an edge is added between two genes if their distance is below the threshold. Hereby, both the choice of the distance function and of the threshold influences the construction of the gene interaction graphs. In Kaba et al. 2007, several criteria are developed to select thresholds in an appropriate way. Here we discuss the impact of the choice of the distance function; our results suggest that this choice does not effect the final decomposition of the gene interaction graphs into clusters.
