Abstract
Metabolomic genome-wide association studies (mGWASs), or metabolomic quantitative trait locus (metQTL) analyses, are gaining growing attention. However, robust methods and analysis guidelines, vital to address the complexity of metabolomic data, remain to be established. Here, we use whole-genome sequencing and metabolomic data from two independent studies to compare different approaches. We adopted three popular data transformation methods for metabolite levels-(i) log(10) transformation, (ii) rank inverse normal transformation, and (iii) a fully adjusted two-step procedure-and compared population-based versus family-based analysis approaches. For validation, we performed permutation-based testing, Huber regression, and independent replication analysis. Simulation studies were used to illustrate the observed differences between data transformations. We demonstrate the advantages and limitations of popular analytic strategies used in mGWASs where especially low-frequency variants in combination with a skewed metabolite measurement distribution can lead to potentially false-positive metQTL findings. We recommend the rank inverse normal transformation or robust test statistics such as in family-based association tests as reliable approaches for mGWASs.