2.4. Data analysis

Feature detection, retention time (RT) correction, and preliminary statistical analyses were carried out using XCMS online (XCMSOnline version 2.2.5; XCMS version 1.47.3; CAMERA version 1.26.0, the Scripps Center for Metabolomics, La Jolla, CA; https://xcmsonline.scripps.edu). Raw data conversion, GC/MS spectra processing, set parameters, etc., were described in Yu et al. (9). All chromatograms were simultaneously analyzed under the same conditions. From seven different conditions of the forced aging procedure (Figure 1), the multi-group job was selected, and the seven groups were classified.

Using the extracted features from XCMS-online software as the Y-Variable, a bivariate Pearson's correlation was performed. The marked significant Y-variables, which were the optimized candidate features based on a false discovery rate≤ 0.05, were selected as candidate features for network analysis and qualitative and quantitative analyses. To further simplify these analyses and delete the corresponding features of the IS and ethanol, the m/z and RT of all extracted features were compared in METLIN (https://metlin.scripps.edu) and the NIST 05 library on the GC/MS workstation (Agilent Technologies Inc., USA).

The identification of candidate features in network analysis directly referred to the existing our study (14): The aligned matrix (/.xlsx) from XCMS-Oline software (https://xcmsonline.scripps.edu) consists of all the exported metabolite features-putative identifications through METLIN standard database matching, which are defined as ions with unique m/z and RT values (Supplementary material). Finally, MS features were validated by (i) crosschecking their presence in the raw data, (ii) comparing their features with those present in the NIST 05 (matching degree %) and NIST 98 MS library (http://webbook.nist.gov), and (iii) comparing with the existing identifications (14–17, 46).

Finally, all identified volatile compounds, including unknowns, were directly quantified using HS-SPME-GC/MS.

Bivariate Pearson's correlation was carried out using IBM SPSS Statistics for Windows, version 19.0 (IBM Corp., Armonk, NY USA). Network analysis of the optimized candidate features was performed with Gephi version 0.9.2 (https://gephi.org/) using the Fruchterman–Reingold algorithm. All box plots and PCA sores plots were exported directly from XCMS-online software. Data trends were graphed with OriginPro 8.5.0 SR1 (The OriginLab Corporation, USA).