Omics studies are now an integral part of systems biology research. Untargeted metabolomics studies result in hundreds or thousands of metabolites measured in blood or other tissue. Among many other applications, measured metabolites are used in biomarker discovery to study human diseases. In this setting, the goal of a metabolomics study is typically to generate promising hypotheses. To ensure replicability of metabolomics findings, it is now standard practice to use training + validation samples. In this approach, discovered significant associations with the disease are validated in an independent dataset which was not used for discovery. It is important therefore to power biomarker discovery studies for this two-stage approach. In this talk we will discuss several approaches that can be used to calculate the power of two-stage analyses of omics data. We use the proportion of non-null markers to estimate the expected number of true positives detected in the training data with the FDR approach to correct for multiple comparisons. We then calculate power to detect 80% of true positives in the test set.