Fig. 4

Positive Predictive Values (PPVs) and progression of DVCs from normal-adjacent tissue in invasive breast cancer. a PPVs of differentially variable CpGs (DVCs) selected by each of five different DV algorithms from comparing 50 normal breast samples from cancer-free women to 42 normal samples adjacent to breast cancers, with the PPVs estimated in 306 invasive breast cancers (compared to same 50 normal reference samples). The number of top-ranked selected DVCs increases along the panels from left to right. PPVs were estimated for hyper-and-hypomethylated DVCs separately: in the case of hypermethylated (hypomethylated) DVCs, PPV was estimated as the fraction of these CpGs attaining a t-statistic larger (lower) than 1.96 (P < 0.05) when comparing invasive cancer to normal. b Left panel: for the top 500 DVCs selected using IEVORA (comparing normal breast samples to normal-adjacent breast tissue), scatterplots compare the DNA methylation values of these sites in the 42 normal adjacent samples (x-axis, NADJ) to the corresponding DNA methylation values in the matched breast cancers (y-axis, BC). Observe how hypermethylated DVCs tend to exhibit further increases in DNAm in the breast cancers that are matched to their corresponding normal-adjacent tissue, whereas the opposite is true for hypomethylated DVCs. Right panel: as left panel, but now plotting the difference in DNAm between the normal-adjacent sample and normals (x-axis,NADJ-N) to the corresponding difference in DNAm between the matched breast cancer and normals (BC-N). We note that because each data point corresponds to 1 CpG site in one patient who provided a normal-adjacent and breast cancer sample, that some of the hypermethylated (hypomethylated) DVCs may exhibit lower (higher) methylation in some of the normal-adjacent samples compared to the normal state