Fang Ma

Angelica sinensis (Oliv.) Diels (AS) is a medicinal and food plant that has long faced a persistent challenge: its quality and price are often influenced by environmental conditions and geographical origins. To achieve substantial profits, items that are not produced in primary regions, along with counterfeit products, are frequently misbranded as originating from main production areas; this leads to fraud regarding geographic origin and product tampering. Rapid, effective and feasible methods for distinguishing the geographic origin of AS are important for ensuring consumer safety and protecting their interests. This study establishes the authenticity and geographical origins of AS. Meanwhile, diverse machine learning strategies are used to identify the optimal combination by incorporating spectral pre-processing techniques, feature wavenumber selection methods and classification algorithms. The findings reveal that the backpropagation neural network (BPNN), convolutional neural network (CNN) and radial basis function neural network (RBF) excel in determining the authenticity of AS. To distinguish among different growing environments of AS, three models obtained 98.94% classification accuracy on the test set: (1) multiplicative scatter correction (MSC) pre-processing with an RBF classifier, (2) standard normalised variate (SNV) pre-processing with an RBF classifier and (3) Savitzky–Golay (SG) smoothing pre-processing, competitive adaptive reweighted sampling (CARS) for selecting features and a BPNN for classification. This study validates the feasibility of ensemble learning combined with MIR for discriminating AS from authenticity and different geographical sources.

The dried roots of the perennial herb Angelica sinensis (Oliv.) Diels (AS) are commonly used as medicinal and edible resources. In commercial planting, early bolting and flowering (EB) of ca. 60% in the medicine formation period reduces root yield and quality, becoming a significant bottleneck in agricultural production. In the cultivation process, summer bolting (SB) occurs from June to July, and autumn bolting (AB) occurs in September. The AB root is often mistaken for the AS root due to its similar morphological characteristics. Few studies have involved whether the root of AB could be used as herbal medicine. This study explored and compared the accumulation dynamics of primary and secondary metabolites in AS and EB roots during the vegetative growth stage (from May to September) by light microscopy, ultraviolet spectrometry, and HPLC methods. Under a microscope, the amount of free starch granules and oil chamber in the AS root increased. On the contrary, they decreased further from EB-Jul to EB-Sep. By comparison, the wall of the xylem vessel was slightly thickened and stacked, and the cell walls of parenchyma and root cortex tissue were thickened in the EB root. Early underground bolting reduces soluble sugar, soluble protein, free amino acids, total C element, total N element, ferulic acid, and ligustilide accumulation, accompanied by the lignification of the root during the vegetative growth stage. Furthermore, a total of 55 root samples from different bolting types of AS root (29 samples), SB root (14 samples), and AB root (12 samples) were collected from Gansu Province during the harvesting period (October). The later the bolting occurred, the less difference there was between unbolted and bolted roots in terms of morphological appearance and efficacy components. Fourier transform infrared spectroscopy with the attenuated total reflection mode (ATR-FTIR) provides a “holistic” spectroscopic fingerprinting of all compositions in the tested sample. The ATR-FTIR spectrum of the AB root was similar to that of the AS root. However, the number and location of absorption peaks in the spectra of SB were different, and only one strong absorption peak at 1021 cm⁻¹ was regarded as the characteristic peak of C-O stretching vibration in lignin. The ATR-FTIR spectra can be effectively differentiated based on their various characteristics using orthogonal partial least squares discrimination analysis (OPLS-DA). Results were assessed using multiple statistical techniques, including Spearman’s correlation, principal component analysis (PCA), partial least squares discriminant analysis (PLS-DA), and OPLS-DA. Among these methods, the ATR-FTIR data demonstrated the most effective outcomes in differentiating between viable and non-viable roots for their application in herbal medicine. Essential substances are ferulic acid and flavonoid, which are much more abundant in the AB root. It provides a material basis for the pharmacological action of the AB roots and a theoretical basis for improving their availability.