Rapid Determination of Polysaccharides in Cistanche Tubulosa Using Near-Infrared Spectroscopy Combined with Machine Learning.

Abstract

Cistanche tubulosa, as a homology of medicine and food, not only has a unique medicinal value but also is widely used in healthcare products. Polysaccharide is one of its important quality indicators. In this study, an analytical model based on near-infrared (NIR) spectroscopy combined with machine learning was established to predict the polysaccharide content of C. tubulosa. The polysaccharide content in the samples determined by the phenol-sulfuric acid method was used as a reference value, and machine learning was applied to relate the spectral information to the reference value. Dividing the samples into a calibration set and a prediction set using the Kennard-Stone algorithm. The model was optimized by various preprocessing methods, including Savitzky-Golay (SG), standard normal variate (SNV), multiple scattering correction (MSC), first-order derivative (FD), second-order derivative (SD), and combinations of them. Variable selection was performed through the successive projections algorithm (SPA) and stability competitive adaptive reweighted sampling (sCARS). Four machine learning models were used to build quantitative models, including the random forest (RF), partial least-squares (PLS), principal component regression (PCR), and support vector machine (SVM). The evaluation indexes of the model were the coefficient of determination (R2), root-mean-square error (RMSE), and residual prediction deviation (RPD). RF performs best among the four machine learning models. R2c （calibration set coefficient of determination） and RMSEC （root mean square error of the calibration set）, %, were 0.9763. and 0.3527 for calibration, respectively. R2p （prediction set coefficient of determination）, RMSEP （root mean square error of the prediction set）, %, and RPD were 0.9230, 0.5130, and 3.33 for prediction, respectively. The results indicate that NIR combined with the RF is an effective method applied to the quality evaluation of the polysaccharides of C. tubulosa. Four quantitative models were developed to predict the polysaccharide content in C. tubulosa, and good results were obtained. The characteristic variables were basically determined by the sCARS algorithm, and the corresponding characteristicgroups were analyzed.