Fig. 4

Multidimensional scaling analysis of variability values and selected features. Homogeneity distribution density was calculated for each csCOG as described in Material and Methods. Classical multidimensional scaling (cmdscale function in R) was applied to visualize the relationship between csCOGs. Hellinger distance (one of the conceptually simplest distance measures which is also symmetrical and metric) was used to quantify the similarity between each two probability distributions. Results for the first two dimensions were used to construct plots. Variability of the data points are shown as follows: Conserved (0–0.5): light blue; medium (0.5–2.0): light gray; variable (> 2.0)” dark blue. The following features are overlayed onto points: presence in the set of core genes—red dots; high gain rate (> 2.5)—magenta dots; membrane (csCOGs with the average fraction of proteins with predicted transmembrane segments > 0.333)—dark green dots; secreted (csCOGs with the average fraction of proteins with signal peptide > 0.333), microsatellite like regions (the average fraction of protein sequences in the csCOG identified >= 0.15)—orange dots; high paralogy (> 2.0)—dark gray dots