Abstract
Both metabolomic and genomic approaches are valuable for risk analysis, however typical approaches evaluating differences in means do not model the changes well. Gene polymorphisms that alter function would appear as distinct populations, or metabotypes, from the predominant one, in which case risk is revealed as changed mixing proportions between control and case samples. Here we validate a model accounting for mixed populations using biomarkers of fatty acid metabolism derived from a case/control study of acute coronary syndrome subjects in which both metabolomic and genomic approaches have been used previously. We first used simulated data to show improved power and sensitivity in the approach compared to classic approaches. We then used the metabolic biomarkers to test for evidence of distinct metabotypes and different proportions among cases and controls. In simulation, our model outperformed all other approaches including Mann-Whitney, t-tests, and χ(2). Using real data, we found distinct metabotypes of six of the seven activities tested, and different mixing proportions in five of the six activity biomarkers: D9D, ELOVL6, ELOVL5, FADS1, and Sprecher pathway chain shortening (SCS). High activity metabotypes of non-essential fatty acids and SCS decreased odds for acute coronary syndrome (ACS), however high activity metabotypes of 20-carbon fatty acid synthesis increased odds. Our study validates an approach that accounts for both metabolomic and genomic theory by demonstrating improved sensitivity and specificity, better performance in real world data, and more straightforward interpretability.