Fig. 1

Informativeness assessment by the proposed omnibus test. a Performance evaluation of the omnibus test compared with naïve tests. Simulated datasets are used to assess type I error (“No dependence” between the p value and the covariate, first column) and power (“Weak dependence” and “Moderate dependence”, second and third column). Varying signal intensities (“Sparse signal,” “Medium signal,” and “Dense signal”, first to third row) as well as different data generation models are studied. “Cont.(Mon.),” “Cont.(Non-Mon),” and “Cat.” represent continuous covariates with a monotonic dependence, continuous covariates with a non-monotonic dependence, and categorical covariates, respectively. Naïve tests refer to Spearman’s rank correlation test and Kruskal-Wallis test for continuous and categorical covariates, respectively. The nominal level for type I error is 0.05 (dashed line). b Informativeness of EWAS-relevant covariates as assessed by the omnibus test. Boxplot depicts the distribution of the omnibus test p values across datasets. The x-axis is on a log scale, and the boxplots are ordered by the median values. The dashed line indicates p value 0.05