Simca 15

SPSS (IBM Corp. version 25.0) and PAST 3 [36 ] software packages were used for descriptive and inferential statistics. Group differences between continuous variables were tested using the Mann–Whitney U test, and categorical variables were tested using the chi-squared test. All tests were two-sided, and correction for multiple comparisons was conducted using the Benjamini–Hochberg false discovery rate (FDR). p-values were considered significant at FDR <0.05. Orthogonal partial least square discriminant analysis (OPLS-DA) (SIMCA 15; Sartorius Stedim Data Analytics AB, Malmö, Sweden) was used to search for differences in oral microbiota between the cases and controls. Clustering of subjects by bacterial taxa in the saliva microbiota was performed by agglomerative hierarchical cluster analysis using Ward’s method, and the linear discriminant analysis (LDA) effect size (LEfSe) method was used for high-dimensional class comparisons [37 (link)]. Potential molecular functions of the saliva microbiota were predicted using the Phylogenetic Investigation of Communities by Reconstruction of Unobserved States (PICRUSt2) plugin for QIIME2 and converted to functions via the Kyoto Encyclopedia of Genes and Genomes (KEGG) Orthology database (https://www.genome.jp/kegg/ko, accessed 1 September 2020) [38 (link)].

This study employed PCA to analyze and interpret the differences observed in the PXRD patterns and their corresponding PDF data. As outlined in the study conducted by Karmwar et al., the preprocessing and scaling of the data were carried out using SIMCA 15 (Ver. 15.0, Sartorius, Göttingen, Germany) [19 (link)]. During the PCA analysis, a specific PDF interatomic distance range was selected. The ranges chosen for PCA analysis were 0–15 Å or 15–100 Å. Subsequently, the PCA data were obtained for each amorphous sample.

Classical statistical analysis was carried out in Statistica 13 (TIBCO Software Inc., Tulsa, OK, United States). The data were examined for normality with Shapiro–Wilk’s test; the majority of the parameters were not normally distributed, and thus, non-parametric statistics were used. Between-group differences in continuous parameters were analyzed with Mann–Whitney U test (two initial groups) or Kruskal–Wallis ANOVA by ranks with post hoc Mann–Whitney U test (more than two initial groups); categorical parameters were analyzed with the maximum-likelihood chi-square test. Within-group differences in continuous parameters were analyzed with Wilcoxon-matched pairs test and categorical parameters with McNemar chi-square test. Effects sizes were calculated for significant differences according to Fritz et al. (2012) (link).
Multivariate data analysis in the form of principal component analysis (PCA) was carried out in SIMCA 15 (Sartorius Stedim Data Analytics AB, Umeå, Sweden). Autofit was used to generate each model; this produces the maximum number of significant components. The components were inspected and excluded if they had eigenvalues <2 or large negative Q² values. The latency, percental frequency, and percental duration parameters were not included in the PCA, and other parameters were excluded when advised by the software due to minimal variance.

The development of chemometric models was undertaken using SIMCA 15 (Sartorius, Sweden). The qualitative models created in SIMCA involved firstly pre-processing the raw data from the FTIR and NIR. This involved the use of Standard Normal Variate (SNV), 1st/2nd Derivative, Savitzky Golay (SG) with 15 points and a quadratic polynomial order along with Pareto scaling. Pre-processing prior to model development allowed focus on the important data points [37 (link)]. PCA, an unsupervised technique, was performed initially, to determine if separate classes could be observed based on the spectral data from both NIR and FTIR for paprika and spent paprika. Following this, a supervised orthogonal partial least squares discriminant analysis (OPLS-DA) model was created to further improve the qualitative models for both NIR and FTIR spectral data. The OPLS-DA algorithm uses both predictive (correlated) and orthogonal (uncorrelated) components to create the classification model to offer a greater understanding of all the aspects of the data. Chemometric analysis was carried out in the range of 550–1800 cm⁻¹ and 2800–4000 cm⁻¹ for FTIR analysis and 4000–9000 cm⁻¹ for NIR analysis. The classes for the binary chemometric models were made up of ‘Paprika’ (n = 104) and ‘Adulterant’ (n = 17).

Data analysis was done using the software Prism 6.0 (GraphPad Software) as indicated in each individual experiment. In summary, statistical difference between two groups was determined using unpaired t test, differences between multiple groups were determined using one-way analysis of variance (ANOVA) with Tukey’s post-test. In this work, a p value below 0.05 was considered significant. Analysis of metabolomic data was done using SIMCA 15 (Sartorius).

The GCxGC-TOF-MS data were exported to Excel^® 2016 and made compatible for chemometrics analysis. All the artifacts and contaminants, such as polymeric materials and stationary phase silicate complexes were excluded from the data in Excel before the data were made using SIMCA-15^® (Sartorius AG, Göttingen, Germany) compatible prior to chemometrics analysis. Principal component analysis (PCA) and orthogonal partial least squares discriminant analysis (OPLS-DA) were performed using SIMCA-15^® software. From the positively correlated observations of the score plot, a column plot was constructed and analyzed, and the contribution of each compound was determined by comparing its weight against that of all the other compounds. A contribution plot for the selected compound displayed the area and the weight t1 (weight = p1p2) on the Y-axis of the plot. Compounds with the most significant weight (t1) were nominated as marker compounds by the model.

Statistical analysis was executed on all continuous variables obtained from LC-MS and GC-MS. To test statistical significance, the Mann-Whitney U-test was implemented in Multiple Array Viewer v.4.9.0 software (http://mev.tm4.org/)[23 (link)]. Multivariate statistical analysis was performed using SIMCA 15 (Sartorius, Göttingen, Germany). Volcano plots were created using GraphPad Prism v.7.04 (GraphPad Software Inc, USA). Bar plots were generated based on LMSstat version 1.0.11 package (https://github.com/CHKim5/LMSstat) in R version 4.0.4 and RStudio version 1.3.1073.