Genome-wide association studies of missing metabolite measures from two population-based studies
TL;DR
It is suggested that missing values in metabolomics are partially non-random and reflect potential genetic variation, and not examined as true missingness due to altered metabolism.
Abstract
Metabolomic analyses are increasingly applied in both etiological and predictive research, but frequently report missing values, which are then either imputed or removed from the analyses, and not examined as true missingness due to altered metabolism. We hypothesized that interindividual genetic variation may account for part of this missingness. We perform a logistic GWAS of metabolite missingness from an untargeted mass spectrometry-based platform in the Netherlands Epidemiology of Obesity Study (N = 594) and the Rhineland Study (N = 4,165). We consider metabolites missing in 10%-90% of individuals in both cohorts (N = 224). GWAS meta-analyses of these metabolites’ probability of missingness revealed 55 metabolome-wide significant associations, including 42 novel ones (p < 1.58 × 10–10), involving 28 metabolites and 41 lead SNPs. Despite considerable pleiotropy, the majority of identified SNP- ‘missing metabolite’ associations are biologically plausible, relating to beta-oxidation, bile acids, steroids, and xenobiotics metabolism. These findings suggest that missing values in metabolomics are partially non-random and reflect potential genetic variation.