5; Bruker, Ettlingen, Germany). Each spectrum was recorded from 4,000 cm−1 to 400 cm−1 using a spectral resolution of 4 cm−1. Signal-to-noise ratio was improved by co-adding 128 interferograms and averaging with the analytical results. Infrared spectra were obtained by subtracting the spectra

of the plates (background) used for deposition of the samples. For multivariate analysis, the digitized original FT-IR spectra were preprocessed (including correction for baseline), and spectral intensity was normalized using the OPUS program (version 6.5; Bruker, Ettlingen, Germany). These preprocessed spectral data were then subjected to multivariate analyses. For multivariate analysis, the 1,800–800-cm−1 region of the FT-IR spectral data rather than the full spectrum was subjected to multivariate analysis.

The preprocessed FT-IR spectral data after a second differentiation were imported into the R statistical analysis program Vorinostat datasheet (version 2.7.2; R Development Core Team) for principal component analysis (PCA), hierarchical clustering analysis (HCA), and partial least squares-discriminant analysis (PLS-DA). PCA as the representative unsupervised pattern recognition method is used to examine the intrinsic variation in the data set, whereas PLS-DA is a supervised pattern recognition method maximizing the separation between samples. PCA and PLS-DA were conducted using the R program. PCA scores extracted from PCA analysis were used to calculate the correlation matrices, and PLS-DA was applied for rapid discrimination among the four ginseng cultivars. To identify variables that Imatinib manufacturer were more valuable for species discrimination among the four ginseng cultivars, we examined PCA loadings. A hierarchical dendrogram was constructed from PLS-DA of the FT-IR data by the unweighted pair–group method with arithmetic mean analysis using the R program; Euclidean distance was used as the similarity measure. A PLS-DA prediction model for cultivar discrimination from the FT-IR spectral data was created by

applying PLS-DA. The PLS-DA model was validated using the cross-validation method, as repeated random subsampling validation Phospholipase D1 [37]. The total dataset was randomly divided into two parts: a training set that was used to build a model (350 samples); and a test set that was not used in the regression model, but was used to verify the model’s predictive ability (130 samples). The classification model for cultivars and cultivation ages of ginseng was developed by a PLS-DA function in the caret package in the R program. A test sample was applied to validate the model. This process was repeated 10 times to reduce error from randomization. The predictive ability of PLS-DA model for prediction of age and cultivar was represented as accuracy and p. As the ginseng plant ages and grows more leaves, typically having five leaflets, development continues until the 5th yr [38]. First-yr ginseng seedlings produced only one compound leaf with three leaflets (Fig. 1A).