4.5. Data Preprocessing and Data Analysis

The raw data files from UHPLC-ESI(+)-MS/MS were exported as mzXML format and uploaded to the XCMS online software (https://xcmsonline.scripps.edu/, accessed date: 10 April 2020). The software was used for feature detection, retention time correction, alignment, automatic integration, and intensity. Parameter settings for XCMS data processing were as follows: Pairwise analysis was performed using centWave for feature detection (Δm/z = 10 ppm, minimum peak width = 5 s, and maximum peak width = 20 s); for retention time correction and chromatograph alignment was performed with an obiwarp method (profStep = 1), minfrac = 0.5, bw = 5, mzwid = 0.015. Statistics analysis was performed with an unpaired parametric t-test (Welch t-test).

The processed data file (CSV format) was exported to MetaboAnalyst 3.0 (www.metaboanalyst.ca, accessed date: 10 April 2020) for the multivariate analysis. Prior features were normalized by sum and scaled by the pareto scaling method. Unsupervised multivariate methods Principal Component Analysis (PCA) was performed to determine differences in metabolic profiles between control and inoculated plants. The Partial Least Squares Discriminant Analysis (PLS-DA) supervised method was used to identify altered metabolites between groups.

The model built from PLS-DA analysis was validated using Leave-one-out cross-validation (LOOCV). The prediction capacity of the model was evaluated by the accuracy parameters R² and Q². Differentially expressed metabolites were selected according to the variable importance in projection (VIP) values ≥1 obtained from the PLS-DA model; p-value ≤ 0.001, from the Welch t-test and the maximum ion intensity ≥10,000.