Figure 1.
Mean absorption spectrum and second derivative spectrum of the data set.
A) Mean infrared absorption spectrum of AC and B) mean second derivative spectrum of AC.
Figure 2.
Scatter plots between optical density of safranin O and different univariate FT-IR PG parameters.
A) Integrated area of carbohydrate region (984–1140 cm−1). B) Integrated area of carbohydrate region normalized with amide I (1584–1720 cm−1). C) Intensity of 2nd derivative peak located at 1374 cm−1. D) Intensity of 2nd derivative peak located at 1062 cm−1. Statistical significance of p<0.001 is indicated by two asterisks (**).
Figure 3.
Root-mean-square error of cross-validation (RMSECV) with different number of components for PLSR and PCR.
Based on RMSECV, 7 components were selected for both models as RMSECV did not decrease significantly with additional components.
Figure 4.
Scatter plots between optical density of safranin O and FT-IR multivariate models.
A) Predicted PG content obtained from PLSR model. B) Predicted PG content obtained from PCR model. Statistical significance of p<0.001 is indicated by two asterisks (**).
Figure 5.
Relative prediction errors of the different FT-IR PG parameters.
A) Integrated area of carbohydrate region (984–1140 cm−1) (mean error: ±35.9%). B) Integrated area of carbohydrate region normalized with amide I (1584–1720 cm−1) (mean error: ±107.8%). C) Intensity of 2nd derivative peak located at 1062 cm−1 (mean error: ±33.0%). D) Intensity of 2nd derivative peak located at 1374 cm−1 (mean error: ±20.6%). E) PLSR model (mean error: ±10.7%). F) PCR model (mean error: ±16.9%). Note that plot B has a different scale on y axis.
Figure 6.
An enzymatically degraded (left) and a normal (right) sample.
Safranin O –stained sections are shown in A and B. Predicted PG contents by the PLSR model are shown in C and D. The loss of superficial PGs is easily seen in C. Depthwise distribution profiles of the same samples analyzed by the PLSR model (black) and by digital densitometry (red) are shown in E and F. The difference between the predicted and measured values are marked with dashed lines in E and F. RMSEs were 0.30 OD and 0.13 OD for the enzymatically modified sample and intact sample, respectively.