With advances in technologies that facilitate metabolome-wide analyses, the incorporation of metabolomics in the pursuit of biomarkers of exposure and effect is rapidly evolving in population health studies. However, many analytic approaches are limited in their capacity to address high-dimensional metabolomics data within an epidemiologic framework, including the highly collinear nature of the metabolites and consideration of confounding variables. In this Children's Health Exposure Analysis Resource (CHEAR) network study, we showcase various analytic approaches that are established as well as novel in the field of metabolomics, including univariate single metabolite models, least absolute shrinkage and selection operator (LASSO), random forest, weighted quantile sum (WQSRS) regression, exploratory factor analysis (EFA), and latent class analysis (LCA). Here, in a Bangladeshi birth cohort (n = 199), we illustrate research questions that can be addressed by each analytic method in the assessment of associations between cord blood metabolites (1H NMR measurements) and birth anthropometric measurements (birth weight and head circumference).
Keywords: CHEAR; Collinearity; Dimension reduction; Feature selection; NMR; Targeted metabolomics.