Metabolomic Profiles Reproducibly Differentiate Heterogeneous Sample Mixtures: A Multi-Lab “Round-Robin” Study

Abstract The clinical utility of metabolomics hinges on reproducibly detecting differences between groups across analytical platforms. A first step toward assuring clinical reliability is establishing the ability to correctly classify heterogeneous mixtures across multiple platforms. No multi-lab studies to date have attempted to use metabolomic profiles to reproducibly cluster heterogeneous mixtures. We evaluated the reproducibility of the TruQuant metabolomics approach in eight labs in the US and UK. Each lab received identical aliquots of 32 samples, containing known mixtures of beef (A), chicken (B), and/or pork (C) extracts of varying homogeneity levels. Metabolomic profiles were generated for each mixture in duplicate using each lab’s LC−MS methods. Anonymized, processed data were analyzed centrally. Despite differences in the number of features detected (560−1727 features), all labs correctly differentiated sample mixture types via unsupervised approaches, with generally strong clustering across labs (NMF-adjusted rand index (ARI) = 1 for all labs except one (ARI = 0.84), unsupervised random forest silhouette score mean = 0.58 ± 0.14). Pearson correlations of suppression-corrected metabolite abundances, compared between lab pairs, were also strong for metabolites that were matched using authentic standards (mean = 0.87 ± 0.12 across all labs). Notably, we evaluated the suitability and impact of different data transformation approaches to make lab-to-lab comparisons. Overall, this “round-robin” study design and data show that metabolomic analyses, with appropriate study designs and data handling, enable reproducible conclusions across labs and platforms.