Evaluation of Different Normalization Strategies for Metabolomics Data Acquired by Gas Chromatography-Mass Spectrometry

Untargeted metabolomics data are highly complex and variable. One of the major challenges in processing them is ensuring that biological interpretations reflect the organism metabolic variability rather than undesirable factors arising from random or systematic experimental errors. Here, nine post-acquisition normalization strategies were evaluated on two gas chromatography-mass spectrometry (GC-MS) metabolomics datasets (cellular and serum). GC-MS data are particularly affected by experimental variability since derivatization steps are required. Thus, the use of internal standards (IS) is indicated, although not mandatory. When an IS is not available, post-acquisition normalization methods help correct for variations. Mean, median, and EigenMS normalization proved quite suitable for both datasets, as evidenced by good sample and quality control grouping and within-group relative log abundance (RLA). EigenMS was chosen due to its improved sample grouping. Strategies that use a sample as a reference factor for normalization did not demonstrate the same efficiency, as they introduced biases into the multivariate models. This could be due to a poor choice of reference sample, which compromises the fit of the data matrix. The results observed in this study demonstrate the importance of testing different normalization methods in any metabolomics study to help obtain more robust and reliable matrices for data inspection and interpretation.