基于NIR定量模型快速预测发酵麸皮中还原糖和可溶性蛋白成分的含量

为建立一种能快速预测发酵麦麸还原性糖和可溶性蛋白含量的定量分析模型，本研究以发酵麦麸为样本，采用3，5-二硝基水杨酸比色法（3，5-dinitrosalicylic acid， DNS）和BCA蛋白浓度测定法分别测定样品的还原性糖和可溶性蛋白含量；利用近红外光谱技术（Near infrared spectrum instrument， NIR）结合偏最小二乘法（Partial least squares regression， PLS），比较预处理方法、最佳波长及主成分因子的决定系数R²，建立发酵麦麸还原性糖和可溶性蛋白含量的NIR快速检测定量模型。结果表明：1）发酵麦麸还原性糖定量模型的预处理方法使用一阶导数 （First derivative， FD）+二阶导数 （Second derivative， SD）+标准正态变换 （Standard normal variate， SNV），光谱范围为908~1 670 nm ，主因子数为7时，模型效果最优，其决定系数Rc²为0.904 8，校正均方根误差（Square error corrected， SEC）为1.576 1，相对分析误差（Relative percentage difference， RPD）为3.240 8；外部验证集决定系数Rp²为0.954 9且还原糖活性成分的验证集样本的测定值与NIR光谱预测值的P值为0.959 5>0.05。2）发酵麦麸可溶性蛋白定量模型的预处理方法使用一阶导数 （First derivative， FD）+二阶导数 （Second derivative， SD）+标准正态变换 （Standard normal variate， SNV），光谱范围为908~1 670 nm ，主因子数为10时，模型效果最优，其决定系数Rc²为0.938 2，校正均方根误差（Square error corrected， SEC）为2.003，相对分析误差（Relative percentage difference， RPD）为4.021 9；外部验证集决定系数Rp²为0.994 4，且可溶性蛋白活性成分的验证集样本测定值与NIR光谱预测值的P值为0.901 9>0.05。综上，建立的NIR光谱定量模型稳定性和准确性较好，且预测准确度良好，可用于快速预测发酵麦麸样品的还原性糖和可溶性蛋白含量。

Rapid prediction of reducing sugar and soluble protein content in fermented bran based on NIR quantitative analysis model

This study aims to establish a quantitative analysis model that can quickly predict the content of reducing sugars and soluble proteins in fermented wheat bran（FWB）. FWB was taken as the study material， and the reducing sugar and soluble protein contents of the samples were determined by the colorimetric method of 3，5-dinitrosalicylic acid （3，5-dinitrosalicylic acid， DNS） and the BCA protein concentration assay （Bicinchoninic Acid Assay， BCA）， respectively. Using Near infrared spectrum instrument （NIR） combined with Partial least squares regression （PLS）， the pre-processing method， the optimal wavelength and the coefficient of determination R² of the principal component factors were compared， and a quantitative model for the rapid detection of NIR in the content of reducing sugars and soluble proteins in FWB were established. The results showed that： 1） The pretreatment method of quantitative model of reduced sugar in fermented wheat bran used First derivative （FD）+ Second derivative （SD）+ Standard normal variate （SNV）. When the spectral range was 908-1 670 nm and the number of principal factors was 7， the model had the best performance. The determination coefficient Rc² was 0.904 8， and the square error corrected （SEC） was 1.576 1. The Relative percentage difference （RPD） was 3.240 8. The determination coefficient Rp² of the external validation set was 0.954 9， and the P value between the measured values of the validation set samples of reducing sugar active ingredients and the predicted values of NIR spectra was 0.959 5>0.05. 2） The pretreatment method for quantitative model of fermented wheat bran soluble protein was First derivative （FD）+ Second derivative （SD）+ Standard normal variate （SNV）. When the spectral range was 908-1 670 nm and the number of principal factors was 10， the model had the best performance. The determination coefficient Rc² was 0.938 2， and the corrected square error corrected （SEC） was 2.003. The Relative percentage difference （RPD） was 4.021 9. The coefficient of determination （Rp²） of the external validation set was 0.994 4， and the P value between the measured values of the validation set samples and the predicted values of the NIR spectra of the soluble active components was 0.901 9>0.05. To sum up， established NIR spectral quantitative model in the study has good stability and accuracy， and can be used to rapidly predict the contents of reducing sugar and soluble protein in FWB samples.