Network-based metabolite ratios for an improved functional characterization of genome-wide association study results

Genome-wide association studies (GWAS) with metabolite ratios as quantitative traits have successfully deepened our understanding of the complex relationship between genetic variants and metabolic phenotypes. Usually all ratio combinations are selected for association tests. However, with more metabolites being detectable, the quadratic increase of the ratio number becomes challenging from a statistical, computational and interpretational point-of-view. Therefore methods which select biologically meaningful ratios are required.\n\nWe here present a network-based approach by selecting only closely connected metabolites in a given metabolic network. The feasibility of this approach was tested on in silico data derived from simulated reaction networks. Especially for small effect sizes, network-based metabolite ratios (NBRs) improved the metabolite-based prediction accuracy of genetically-influenced reactions compared to the all ratios approach. Evaluating the NBR approach on published GWAS association results, we compared reported all ratio-SNP hits with results obtained by selecting only NBRs as candidates for association tests. Input networks for NBR selection were derived from public pathway databases or reconstructed from metabolomics data. NBR-candidates covered more than 80% of all significant ratio-SNP associations and we could replicate 7 out of 10 new associations predicted by the NBR approach.\n\nIn this study we evaluated a network-based approach to select biologically meaningful metabolite ratios as quantitative traits in GWAS. Taking metabolic network information into account facilitated the analysis and the biochemical interpretation of metabolite-gene association results. For upcoming studies, for instance with case-control design, large-scale metabolomics data and small sample numbers, the analysis of all possible metabolite ratios is not feasible due to the correction for multiple testing. Here our NBR approach increases the statistical power and lowers computational demands, allowing for a better understanding of the complex interplay between individual phenotypes, genetics and metabolic profiles.