Figure 2

Identification of the optimal cutoff for calling a gap between two DMRs. (A) Histogram of the log2 distance of the nearest CpGs in Sample 1. A spike at zero log2 base pairs distance represents the reverse complement of CpGs (GpC) on the other strand. (B) Bimodal normal distribution fitting on the log2 distance of adjacent CpGs genome-wide in AML sample 1. Two distributions (red, and green) are shown that account for two separate data densities (dotted line). (C) Weighted sum of penalty changes (blue line) over log2 distances. The red line is the optimized log2 DMR distance with the lowest weighted penalty from the cost function