Abstract
Metabolomics commonly relies on using one-dimensional (1D) (1)H NMR spectroscopy or liquid chromatography-mass spectrometry (LC-MS) to derive scientific insights from large collections of biological samples. NMR and MS approaches to metabolomics require, among other issues, a data processing pipeline. Quantitative assessment of the performance of these software platforms is challenged by a lack of standardized data sets with "known" outcomes. To resolve this issue, we created a novel simulated LC-MS data set with known peak locations and intensities, defined metabolite differences between groups (i.e., fold change > 2, coefficient of variation ≤ 25%), and different amounts of added Gaussian noise (0, 5, or 10%) and missing features (0, 10, or 20%). This data set was developed to improve benchmarking of existing LC-MS metabolomics software and to validate the updated version of our MVAPACK software, which added gas chromatography-MS and LC-MS functionality to its existing 1D and two-dimensional NMR data processing capabilities. We also included two experimental LC-MS data sets acquired from a standard mixture andMycobacterium smegmatiscell lysates since a simulated data set alone may not capture all the unique characteristics and variability of real spectra needed to assess software performance properly. Our simulated and experimental LC-MS data sets were processed with the MS-DIAL and XCMSOnline software packages and our MVAPACK toolkit to showcase the utility of our data sets to benchmark MVAPACK against community standards. Our results demonstrate the enhanced objectivity and clarity of software assessment that can be achieved when both simulated and experimental data are employed since distinctly different software performances were observed with the simulated and experimental LC-MS data sets. We also demonstrate that the performance of MVAPACK is equivalent to or exceeds existing LC-MS software programs while providing a single platform for processing and analyzing both NMR and MS data sets.