Simca

After pretreatment, 5,438 detected proteins were retained. Data were processed with Proteome Discoverer software (Thermo Fisher Scientific, version 2.4.0.305). Principal component analysis (PCA) was performed using the R package. (version3.6.3, https://www.r-project.org/) or SIMCA (V16.0.2, Sartorius Stedim Data Analytics AB, Umea, Sweden). Differential expressed proteins (DEPs) were defined as student’s t-test p-value < 0.05 and fold change ≤ 0.83 or fold change ≥ 1.2. DEPs were visualized in the form of volcano plot. Hierarchical clustering for representing the DEPs was conducted by R Package pheatmap. The eukaryotic clusters of orthologous groups (KOG) database (http://www.ncbi.nlm.nih.gov/COG/) of protein database was carried out for functional classification of DEPs. Gene ontology (GO) database (http://geneontology.org/) was used to classify and annotate the functions of differentially expressed proteins. All enrichment analyses were based on the Fisher’s exact test with Benjamini−Hochberg correction (p < 0.05).

The metabolomics and proteomics data, after a batch-effect adjustment and log transformation, were analysed using multivariate data analysis software SIMCA (version 16, Sartorius Stedim Biotech, Umeå, Sweden) and MetaboAnalyst 4.0 [29 (link)]. The Gene Ontology Resource and Enrichr were used for Enrichment analysis. The cut-off level for significant metabolites was a signal-to-noise (S/N) ratio of 10, while for proteins, it was a relative abundance of 1 × 10⁵. For statistical analysis of both metabolome and proteome, a fold change of ≤0.5 (downregulation) or ≥2.0 (upregulation), and a Benjamini–Hochberg adjusted p-value of ≤0.05 was used. Metabolomic and proteomic outputs were integrated using the ‘Joint-pathway analysis tool’ of MetaboAnalyst 4.0 and Paintomics 3 [30 (link)], and the false discovery rate (FDR) was used to report adjusted p-values.

All data were analysed using one-way ANOVA and followed up with Fisher’s LSD multiple comparison test with the confidence level set at p < 0.05. End-point of the stationary phase of growth was determined by three consecutive points that are not significantly different from each other. Analysis was carried out using SPSS software (IBM SPSS Statistics 26.0).
Multivariate analysis was applied to analyse FAME output from GC-FID. For the first stage of experiment, comparisons of FAME data between different salinities were analysed using Principal Component Analysis (PCA). In the second stage of experiment, variation of FAME within different culture days was analysed using Partial Least Squares Discriminant Analysis (PLS-DA). All sample values were linear-logged and Pareto-scaled. Analysis was carried out in SIMCA (16.0 Umetrics; Sartorius) software.

Relative quantification was made by expressing the molar amount of each individual lipid as a percentage of the sum of lipids analyzed in the sample. All samples were analyzed in duplicate injections and the average was used. Relative standard deviation for duplicates had an average of 2.4% (range 0–14%) and QC samples were distributed throughout the run to assess method stability during the run. Data was log transformed and was provided as geometric mean and standard deviation. Repeated measures ANOVA was used for the comparison between visits. Multivariate analysis was made using the SIMCA software (SIMCA, Sartorius AG, Germany). Difference between sampling method were evaluated by Orthogonal Partial Least Squares Discriminant Analysis (OPLS-DA). The OPLS-DA model cross validation was based on a grouping by subject id, so that all data from one subject was removed from the model and then predicted by the model. The coefficients of variation (CV) for repeated PEx measurements were calculated separately for each lipid. For this, the standard deviation of all subjects was divided by the mean value for all subjects. As the CVs were not normally distributed, we provided their medians and IQR for all lipids in the results. A value below 10 can be considered as good. No correction for multiple testing was done due to the exploratory nature of our study.

Data from the crude HD samples were
initially analyzed by principal component analysis (PCA) to get an
overview of the variation in the data and detect potential outliers.
Subsequent analyses were performed in parallel using two machine learning
algorithms: orthogonal partial least squares discriminant analysis
(OPLS-DA)^{36 (link)} and random forest (RF),^{37 (link)} which were performed in Simca (version 15, Sartorius
Stedim Biotech, Germany) and the R software (R³⁸ version 4.0.3 with a random forest 4.6–14³⁹ package), respectively.

The statistical and Spearman’s correlation analyses were performed using the Statistical Package for the Social Sciences (SPSS 19.0, IBM, Inc., Armonk, NY, USA), and the significances among the groups were evaluated using the multiple comparative analysis of variance (ANOVA). The microbial data were analyzed on the online platform of Majorbio Cloud Platform (www.majorbio.com (accessed on 1 May 2022)). The major flavor compounds were screened and analyzed by the principal component analysis (PCA), O2PLS-DA, and permutations plot using SIMCA^® (version 14.1, Sartorius Stedim Data Analytics AB, Umeå, Sweden). The results of a Spearman’s correlation analysis indicated that the relationship between the micro-organisms and flavor substances was established by Gephi (0.9.25) (https://gephi.org/ (accessed on 1 May 2022)), and the network map was optimized using the Cytoscape (3.9.1) software (http://www.cytoscape.org/ (accessed on 1 May 2022)). All the experiments were performed in triplicate, and the data are expressed as the means ± standard deviations.

Significant differences between means were analyzed by one-way ANOVA using SPSS (version 26.0, SPSS Inc., Chicago, IL, USA). The score scatter plot and permutation plot for the OPLS-DA were performed by SIMCA (V14.1, Sartorius Stedim Data Analytics AB, Umea, Sweden). HCA (hierarchical clustering analysis) was drawn using an integrative toolkit—TBtools. The bar charts were generated using Origin 2018 (OriginLab, Northampton, MA, USA).