Fig. 4

The empirical power of 2dFDR on real omics datasets. 2dFDR made more discoveries than 1dFDR-A across FDR target levels for A TCGA hepatocellular carcinoma transcriptomics (RNA-Seq) dataset (m = 19,329, confounder: gender and ethnicity), B insulin resistance metabolomics dataset pooling polar metabolites and molecular lipids (m = 1201, confounder: BMI). C, D Evaluation of 2dFDR on 54 epigenomics (450 K methylation array) datasets from EWAS of various phenotypes (m ≅ 450,000, confounder: cell mixtures). C Boxplot comparing the number of DMPs (differentially methylated positions) detected by 2dFDR and 1dFDR-A at 5% FDR over the 54 datasets. 2dFDR recovered more DMPs than 1dFDR-A in 43 datasets. Each gray dot represents a dataset, and the same dataset is connected by a line. D The distribution of the percent increase in detection power over the 54 datasets. 2dFDR achieves a median percent increase of 136% over 1dFDR-A