2.1.3 Metabolomics Data Analysis

The raw data were analyzed using MassHunter Profinder 8.0 and MassProfiler Professional (MPP) 15.1 software (Agilent). Metabolite structures were identified using an in-house annotated metabolite database created using MassHunter PCDL manager 8.0 (Agilent) based on monoisotopic neutral masses (<5 ppm mass accuracy) and chromatographic retention times. A molecular formula generator (MFG) algorithm in MPP was used, based on weighted consideration of monoisotopic mass accuracy, isotope abundance ratios, and spacing between isotope peaks. A tentative compound ID was assigned when MFG scores concurred with the PCDL database for a given candidate molecule. Tentatively assigned molecules were verified based on a match of LC retention times and/or MS-MS fragmentation spectra for pure molecule standards contained in the in-house database. Common R packages such as omu (https://cran.r-project.org/web/packages/omu) and Metaboanalyst (https://www.metaboanalyst.ca) were used for data analysis and visualization, including hierarchical clustering, pathway enrichment analysis, principal component, and partial least squares-discriminant analyses. For analysis using Metaboanalyst, non-informative features were filtered out using the interquantile range method. Missing values were estimated feature-wise using a k-nearest neighbors (KNN) algorithm. Samples were normalized by median, and the dataset was log-transformed and mean-centered.