Rapid Identification of Medicinal Polygonatum Species and Predictive of Polysaccharides Using ATR‐FTIR Spectroscopy Combined With Multivariate Analysis

ABSTRACT Introduction Medicinal Polygonatum species is a widely used traditional Chinese medicine with high nutritional value, known for its anti‐fatigue properties, enhancement of immunity, delays aging, improves sleep, and other health benefits. However, the efficacy of different species varies, making the quality control of medicinal Polygonatum species increasingly important. Polysaccharides are important in medicinal Polygonatum species because of their potential functional properties, such as antioxidation, hypoglycemia, protection of intestinal health, and minimal toxicological effects on human health, as well as high polysaccharide levels. Objective This study developed a qualitative medicinal Polygonatum species model and a polysaccharides predictive model based on attenuated total reflection Fourier transform infrared spectroscopy (ATR‐FTIR) combined with a multivariate analysis approach. Materials and Methods ATR‐FTIR spectral information of 334 medicinal Polygonatum species samples was collected and the spectral information of different modes was analyzed. The ATR‐FTIR spectral differences of three medicinal Polygonatum species were studied by multivariate analysis combined with four spectral preprocessing and three variable selection methods. For the prediction of polysaccharides in Polygonatum kingianum Collett & Hemsl. (PK), we initially determined the actual content of 110 PK polysaccharide samples using the anthrone‐sulfuric acid method, then established partial least squares regression (PLSR) and kernel PLSR models in conjunction with attenuated total reflectance Fourier transform infrared (ATR‐FTIR) spectroscopy. Results In the visualization analysis, the orthogonal partial least squares‐discriminant analysis (OPLS‐DA) model based on second‐order derivative (SD) preprocessing was most suitable for medicinal Polygonatum species species binary classification, spectral differences between Polygonatum cyrtonema Hua (PC) and other species are evident; in the hard modeling, SD preprocessing improves the accuracy of non‐deep learning models for the classification of three medicinal Polygonatum species. In contrast, residual neural network (ResNet) models were the best choice for species identification without preprocessing and variable selection. In addition, the partial least squares regression (PLSR) model and Kernel‐PLSR model can quickly predict PK polysaccharides content, among them, the Kernel‐PLSR model with SD pretreatment has the best prediction performance, residual prediction deviation (RPD) = 7.2870, Rp = 0.9905. Conclusion In this study, we employed ATR‐FTIR spectroscopy and various treatments to discern different medicinal Polygonatum species. We also evaluated the effects of preprocessing methods and variable selection on the prediction of PK polysaccharides by PLSR and Kernel‐PLSR models. Among them, the ResNet model can achieve 100% correct classification of medicinal Polygonatum species without complex spectral preprocessing. Furthermore, the Kernel‐PLSR model based on SD‐ATR‐FTIR spectra had the best performance in polysaccharides prediction. In summary, by integrating ATR‐FTIR spectroscopy with multivariate analysis, this research accomplished the classification of medicinal Polygonatum species and the prediction of polysaccharides. The methodology offers the benefits of speed, environmental sustainability, and precision, highlighting its significant potential for practical applications. In future research, on the one hand, it can be further investigated using a portable infrared spectrometer, and on the other hand, infrared spectroscopy can also be applied to the prediction of other chemical components of medicinal Polygonatum species.