Comparative performance of Chenomx and MagMet in NMR-based metabolite profiling with validation by liquid...

In this work, two computational approaches for metabolite quantification in serum samples using ¹H NMR spectroscopy were evaluated: the spectral matching method (M_SM) implemented in MagMet and the non-linear least squares method (M_NLLS) implemented in Chenomx. The comparison focused on their underlying methodologies, including deconvolution algorithms and user workflows, to assess their relative performance and suitability for metabolomics data analysis. As various analyses (e.g. pattern recognition, classification, biomarker discovery, and pathway analysis) rely on the precision and consistency of input features (e.g., metabolite concentrations), selecting a robust quantification method is essential. Variability in quantification can introduce noise and impact the stability and comparability of analytical outputs. To validate performance, M_SM (MagMet) and M_NLLS (Chenomx) were benchmarked against quantitative NMR (qNMR) and liquid chromatography-tandem mass spectrometry (LC-MS/MS), the latter serving as the primary reference due to its high sensitivity and broad metabolite coverage (Gika et al., 2014) [1]. Although LC-MS/MS may be affected by matrix effects and ion suppression; these factors are well characterized and routinely mitigated through isotope-labeled internal standards and validated analytical workflows. Moreover, LC-MS offers substantially higher sensitivity than NMR, typically by two to three orders of magnitude, enabling the detection and quantification of hundreds to thousands of metabolites within a single analysis (Nagana Gowda and Raftery, 2022) [2]. qNMR was included as a complementary technique to provide orthogonal validation rather than serving as the sole benchmark. Ten independent serum control samples from a healthy reference group were analyzed to account for natural biological variability, enhancing the generalizability of the findings. The comparison was structured around four criteria: (i) quantitative performance, (ii) computational stability, (iii) usability and processing time, and (iv) method-based similarity via partial least squares-discriminant analysis (PLS-DA). This work differs from prior studies by integrating statistical validation, repeatability testing, and practical usability assessment, and by benchmarking computational quantification pipelines against experimentally grounded methods such as qNMR and LC-MS/MS [3-5]. The selected approach is expected to demonstrate improved consistency in quantification relative to the alternative, contributing to more reliable biological interpretations and more reproducible analytical outcomes across datasets.