Technical report: A comprehensive comparison between different quantification versions of Nightingale Health’s 1 H-NMR metabolomics platform

medRxiv

Abstract

Abstract 1 H-NMR metabolomics data is increasingly used to track various aspects of health and disease. With the availability of larger data resources and continuously improving learning algorithms Nightingale Health has recently updated the quantification and calibration strategy of their platform to further align their reported analytes with clinical standards. Such updates, however, might influence backward replicability and could hamper comparison of repeated measures in longitudinal studies. Based on data of the BBMRI.nl consortium (>25.000 samples across 28 studies), we compared Nightingale data, as originally released in 2014 and 2016, with a re-quantified version of this data released in 2020, of which both versions were based on the same original NMR spectra. Apart from 2 discontinued, and 23 newly defined analytes, we overall observe a high concordance between quantification versions, with 73 out of 222 (33%) showing a mean correlation > 0.9 across the 28 Dutch cohorts. Nevertheless, five metabolites consistently showed relatively low correlations (R<0.7) between platform versions, namely acetoacetate (acace), LDL particle size (ldl_d) , saturated fatty acids percentage (sfa_fa) , S-HDL-C (s_hdl_c) and sphingomyelins ( sm) . Previously trained multi-analyte scores, such as our previously published health predictors MetaboAge or MetaboHealth , might be particularly sensitive to platform changes. Whereas the MetaboHealth score replicated well between platform versions, the MetaboAge score indeed had to be retrained due to discontinued metabolites. Notably, both scores projected on the 2020 re-quantified data did recapitulate the original mortality associations observed in the previous version of the data. Concluding, we urge caution when utilizing data from different quantification versions to avoid mixing analytes capturing different underlying aspects of the NMR spectra, having different units, or simply being discontinued.

Authors 5

1. Daniele Bizzarri Aachen Leiden Computational Biology Center Molecular Epidemiology

   Leiden University Medical Center · Delft University of Technology

   Affiliation as printed

   Delft Bioinformatics Lab, TU Delft, Delft, The Netherlands

   Leiden Computational Biology Center, LUMC, Leiden, The Netherlands

   Molecular Epidemiology, LUMC, Leiden, The Netherlands

2. Marcel J. T. Reinders Aachen Leiden Computational Biology Center

   Leiden University Medical Center · Delft University of Technology

   Affiliation as printed

   Delft Bioinformatics Lab, TU Delft, Delft, The Netherlands

   Leiden Computational Biology Center, LUMC, Leiden, The Netherlands

3. Marian Beekman Aachen Molecular Epidemiology

   Leiden University Medical Center

   Affiliation as printed

   Molecular Epidemiology, LUMC, Leiden, The Netherlands

4. P. Eline Slagboom Aachen Molecular Epidemiology Leiden Computational Biology Center

   Leiden University Medical Center · Max Planck Institute for Biology of Ageing · Delft University of Technology

   Affiliation as printed

   Max Planck Institute for the Biology of Ageing, Cologne, Germany

   Molecular Epidemiology, LUMC, Leiden, The Netherlands

   Delft Bioinformatics Lab, TU Delft, Delft, The Netherlands

   Leiden Computational Biology Center, LUMC, Leiden, The Netherlands

5. Erik B. van den Akker corresponding Aachen Leiden Computational Biology Center Molecular Epidemiology

   Leiden University Medical Center · Delft University of Technology

   Affiliation as printed

   Delft Bioinformatics Lab, TU Delft, Delft, The Netherlands

   Leiden Computational Biology Center, LUMC, Leiden, The Netherlands

   Molecular Epidemiology, LUMC, Leiden, The Netherlands