Fig. 6

The dynamics of DV in carcinogenesis and operating characteristics of DV algorithms. a Progression of DV in cervical carcinogenesis. Left panel depicts the DNAm beta-value of a specific CpG (cg10141715) across different disease stages in cervical carcinogenesis, including cytologically normal cells which remain normal 3 years later (N- > N), cytologically normal cells which progress to CIN2+ 3 years later (N- > CIN2+), cervical intraepithelial neoplasia of grade 2 or higher (CIN2+) and cervical cancer. Right panel is a boxplot representation, indicating the P-values from a t-test (TT), Wilcoxon rank sum test (WT) and Bartlett’s test (BT) between the normal state (N- > N) and each of the other 3 stages. b Relative fractions of type-1a, type-1b and type-2 DVCs in cervical and breast carcinogenesis. DVCs were selected using an FDR threshold of 0.05 on the Bartlett’s test P-value. They were defined to be of type-2 if the t-test P-value was not significant (P > 0.05). They were defined to be of type-1 if the t-test P-value < 0.05, and of type-1a if the t-test P-value was more significant than the one from the Bartlett’s test, otherwise defined as type1-b. In the context of the cervix, the reference samples were normal cervical samples from the corresponding study. In the context of breast, the reference samples were normal breast tissue samples from healthy women. c Left panel: Barplots of estimated sensitivity (SE) values averaged over 100 simulated runs for a number of different DV algorithms (standard deviations were small and not shown for convenience). DVCs were selected at an estimated FDR < 0.05. Shown are the overall sensitivities to detect any DVC, and the corresponding sensitivities to detect particular types of DV. Right panel: Boxplots of the true FDRs for each DV algorithm. Green dashed line indicates the line FDR = 0.05