Controlling bias and inflation in epigenome- and transcriptome-wide association studies using the empirical null distribution

Table 2 Bias and inflation correction after adjustment for confounding factors yields optimal power

Mean and standard deviation of the number of false positives and true positives (power) for a simulation study repeated 100×. Data were generated according to the simulation setup of Wang et al. [17]. The table summarizes the results for the naive approach of no adjustment for confounding factors and adjusting for confounding factors using CATE. Both in combination with different approaches are used to control for inflation (and bias): no correction, correction using genomic control, correction using the median and median absolute deviation (MAD), calibration [17], and using our Bayesian method. As a comparison the oracle method is shown where the simulated confounding factors have been added to the linear model

ISSN: 1474-760X