Figure 2

Learning procedures for new emission functions. (a) Bimodality of genome-wide methylation level distributions. Real bisulfite-seq data from [10] are shown. (b) Learning parameters for beta mixtures. Genome-wide methylation level distributions are fitted by beta mixtures, and differential methylation states are represented as alterations between two mixture components. The histograms are depicted using all genomic CpG sites, while the probability distributions are estimated using 10000 randomly selected CpG sites.