Data analysis

Peak intensities of the identified metabolites were normalized using vector and median normalization methods. Relative peak intensities were log-transformed and scaled using Pareto scaling in the MetaboAnalyst program (https://www.metaboanalyst.ca; accessed 18 Aug 2019) (Chong et al. 2018). For principal component analysis (PCA) and hierarchical cluster analysis (HCA) of the metabolite data, the pre-processed and normalized dataset was imported into the ClustVis web tool (https://www.biit.cs.ut.ee/clustvis; accessed 20 Aug 2019) and further analyzed (Metsalu and Vilo, 2015). One-way analysis of variance (ANOVA) followed by post-hoc analysis using Fisher’s least significant difference (LSD) test was used for analysis of statistical significance. The alluvial plot was generated using RAWGraphs (https://rawgraphs.io/; accessed 07 Sep 2019) (Mauri et al. 2017). NetworkAnalyst software (https://www.networkanalyst.ca; accessed 20 Aug 2019) (Zhou et al. 2019) in conjunction with MetaboAnalyst was used to generate and visualize the metabolite-metabolite interaction network. The associations for the metabolite networks were extracted from STITCH (a feature of NetworkAnalyst) and only highly confident interactions were included. Receiver operating characteristic (ROC) curves and the areas under the curve (AUC) were generated using MetaboAnalyst.