Two local dissimilarity measures for weighted graphs with application to protein interaction networks

We extend the Czekanowski-Dice dissimilarity measure, classically used to cluster the vertices of unweighted graphs, to weighted ones. The first proposed formula corresponds to edges weighted by a probability of existence. The second one is adapted to edges weighted by intensity or strength. We show on simulated graphs that the class identification process is improved by computing weighted compared to unweighted edges. Finally, an application to a drosophila protein network illustrates the fact that using these new formulas improves the ’biological accuracy’ of partitioning.