Fig. 1
Application of Jensen-Shannon divergence to DNA methylation. a Geometric explanation of Jensen-Shannon divergence (JSD) in terms of Shannon entropy. The graph shows the entropy of a binary distribution (with probabilities p and q=1−p) in terms of p. For the purpose of illustration, the red dots designate the coordinates of three distributions, P1,P2,P3, that define a red curve segment and a corresponding polygon (red triangle). The curve segment constrains the entropy of the mixture distribution, H〈P〉, while the polygon constrains the corresponding average entropy, 〈H〉. The exact location of both JSD terms depends on the weights. For equally weighted distributions (here πi=1/3 for all i), their locations are given by the blue dots. The corresponding distance (length of the dashed line) equals JSD. b Phase plane in terms of weighted average methylation MET (μ) and diversity index JSD. The methylation state of a cytosine in the population is represented by a point at or below the graph. Four regions of interest are highlighted: three regions with JSD below ≈0.7, LMC (low-methylated cytosines), MMC (medium-methylated cytosines), and HMC (high-methylated cytosines); a region with high JSD for MSCs (metastable cytosines). c Overview of methylome data sources. Four hundred eighty-two Arabidopsis thaliana methylomes from 75 different studies have been analyzed. All methylomes derive from wild-type plants of the Columbia 0 (Col-0) accession. The image is modified based on [26]