Fig. 4

Benchmark on simulated data and GIAB dataset. A Number of segments identified as full match, meaning that both the true start and end changepoint have been accurately identified, partial match, meaning that one of either the true start- or end-changepoint has been accurately identified, or no match, meaning that neither the true start nor end changepoint have been accurately identified, in the high-coverage simulated dataset. For the purpose of this graphic, a changepoint is considered accurately identified if a predicted segment breakpoint is no more than 5% of its containing segment’s length away from the true (ground truth) changepoint. Additional changepoints are the number of predicted changepoints not counting towards a partial or full match (oversegmentation). B Performance of DMR calling on simulated data using the pycoMeth coarse segmentation and varying DMR tests. Each bar group corresponds to the respective interpretation of the y-axis: recall as a measure of test power, precision as a measure of false discovery, and F1-score (harmonic mean of recall and precision). C DMR calling performance on simulated data for varying segmentations. All settings use a Fisher’s exact test for DMR testing. MethCP-pycoMeth and methylKit settings use pycoMeth for DMR testing. D Allele-specific methylation called on HG002. PycoMeth calls were performed using the Fisher exact test with IHW. Colors represent the effect size of the containing segment (absolute differential methylation rate). E DMRs identified between HG003 and HG004