ABCD-DNA outperforms CNV-aware Batman. ROC curves (sensitivity versus 1 − specificity) for three pairwise comparisons are shown for a MeDIP-seq data set (Feber et al. 2011), where Illumina HumanMethylation 27k data is used as an independent source of truly and nontruly differentially methylated regions. “Batman” refers to the CNV-adjusted read densities before running the Batman algorithm and taking differences in methylation estimates. “Naive” refers to a count-based analysis, without accounting for CNV. “ABCD-DNA” refers to a count-based analysis, with additional offsets to account for CNV (estimated from Affymetrix SNP 6.0 data using the PICNIC algorithm). Comparisons are as follows: (A) cancer versus normal; (B) cancer versus benign; and (C) benign versus normal.
