Skip to main content
. 2013 Oct 10;9(10):e1003268. doi: 10.1371/journal.pcbi.1003268

Figure 6. Community structures vary across HVRs.

Figure 6

(A) Variation of information (VI) measures the distance between two different partitions on the same set of nodes (two different sets of community assignments). For a given HVR we compute pairwise VI distances to recovered communities of other HVRs (colored symbols), cys/PoLV (+), UPS classifications (x), and to those in a null model (grey open symbols). Uncertainty in VI measurements is discussed in detail in Text S4. (B) HVRs 1 and 6 are close to each other, indicating that their communities strongly match each other. Histograms show the distributions of VI distances for one partition and 10,000 randomizations of the other. Top, the measured distance between HVRs 5 and 9 falls within the distribution of randomizations, indicated by the arrow. Bottom, the measured distance between HVRs 1 and 6 falls well outside the distribution of randomizations, indicated by the left arrow. For contrast, the silhouette of the top histogram is reproduced. (C) Most HVRs show moderate but positive levels of assortativity [45], the tendency for nodes with similar labels or values to be connected. Assortativity varies by HVR and by label (symbols). For all cases except DBLα classification (DBLα0, DBLα1, DBLα2), assortativity was significantly higher than expected by chance. Solid lines with whiskers show mean assortativity ± one standard deviation for 10,000 randomizations of labels. Z-scores may be found in Figure S5A.