Abstract:
|
Illumina HumanMethyation BeadArrays have become the preferred platform for large-scale studies of DNA methylation (DNAm). In designing a DNAm study, researchers are confronted with many design decisions; chief among them is whether the planned study is well-powered. Power for detecting differential methylation (DM) is affected by the fraction of DM CpGs, effect size distribution, distribution of DNAm, number of replicates, and variation between replicates. The complexities of DNAm array data make assessment of statistical power particularly challenging. To overcome these issues, we propose prospective power assessment using simulations by making assumptions on the aforementioned factors. First, a semi-parametric approach is used to generate data based on actual DNAm datasets. Second, available R packages for DM evaluation are utilized and power is assessed empirically by averaging over simulations. We report the marginal type I error rate, marginal FDR, power, marginal target power, and false discovery cost based on user-specified parameter values. In this presentation, we describe our approach and the ongoing development of an R package that implements the proposed framework.
|