Figure 2

A) PPV comparison between M-values and β-values in the small sample size limit and in the diagnostic setting. Evaluation set sizes were the top 200, 1500, 2500, 1500, and 3000 diagnosis-associated-CpGs for UKOPS, ENDOM, CERVX, BC and LC, respectively, ranked according to regularized t-statistics. Sample sizes were 3 per phenotype (UKOPS), and 2 per phenotype (all others). B) As A) but now in the large sample size limit where 50% training/test partitions were used to estimate the PPV. In both A) and B), the boxplots represent the distribution of the PPV over 50 random training/test set partitions. Wilcoxon rank sum test P-values are given.