Fig. 2
From: Robust, scalable, and informative clustering for diverse biological networks
Recovering disjoint ground-truth clusters from LFR benchmark networks. A Recovery of original LFR communities, by several clustering methods, quantified by normalized mutual information (NMI), grouped by levels of cluster cross-linking (“μ”, x-axis). Cross-linking edge weights are co-varied with frequency of cross-linking for all results in this panel. B Recovery of original LFR communities, quantified by adjusted rand index (ARI). C Recovery of original LFR communities, by several clustering methods, quantified by normalized mutual information, for all networks with the weight distribution exponent (β) level equal to 1.5. D Recovery of original LFR communities, by several clustering methods, quantified by normalized mutual information, for all networks with β level equal 3. E Classic Newman’s modularity of inferred clusters. F Comparison of modularity of inferred clusters relative to the modularity of the ground truth solution. Modularity of inferred solutions may be greater than that of the true clusters (percent difference greater than zero) in noisy (high μ) networks. G Modularity density (“Qds”) of inferred clusters. I Comparison of modularity density of inferred clusters to that of the ground truth solution