Fig. 5

Validation of 2dFDR using EWAS datasets. A The distribution of the confounder-adjusted p values for those 2dFDR-exclusive DMPs in other age EWAS datasets. The diagonal parts show the densities of p values of all loci for the five age EWAS datasets, serving as a baseline. The off-diagonal parts show the distribution of p values of 2dFDR-exclusive DMPs in other age datasets. For instance, the first row shows the distribution of p values for the 2dFDR-exclusive DMPs from EWAS26 in EWAS27 (green), EWAS30 (purple), EWAS39 (brown), and EWAS45 (orange). The distribution is enriched in smaller p values than the distribution of all p values for the respective dataset (diagonal). B Similar validation using two systemic lupus erythematosus (SLE) EWAS datasets (EWAS28 and EWAS29, Additional file 2: Table S1). C Validation using a downsampling strategy on EWAS22 (n = 111). First, a list of “gold standard” DMPs (gDMPs) was created using Bonferroni correction based on the p values from the adjusted analysis on the full dataset. Next, the full dataset was downsampled to smaller sample sizes, and the ability of 2dFDR and 1dFDR-A in recovering those gDMPs was compared. For each sample size, 100 replications were performed, and means and standard errors are plotted. 2dFDR is more powerful in identifying these gDMPs at smaller sample sizes