Authentication of Linderae Radix through plant metabolomics coupled with a machine learning-enhanced in situ hyperspectral imaging approach

Linderae Radix, a medicinally significant herb with a history of over 2,000 years, is highly esteemed for its potential to promote longevity. Derived from the tuberous roots of Lindera aggregata ( L. aggregata ), it encounters difficulties in being distinguished from non-medicinal parts, such as non-fusiform taproots and old roots in the herbal drug market. To address the problem, this study developed a new strategy that integrates non-targeted plant metabolomics with a machine learning-enhanced hyperspectral imaging (HSI) approach for in situ quality assessment. Firstly, a comprehensive metabolomics analysis was conducted using ultra-performance liquid chromatography-quadrupole time-of-flight mass spectrometry (UPLC-QTOF-MS) and gas chromatography-mass spectrometry (GC-MS) to identify 25 and 48 differential metabolites, respectively. Then, combined with machine learning algorithms, HSI in the 400–1,000 nm band achieved visual in situ assessment of different types of L. aggregata roots. Second derivative (2 nd D)-Savitzky-Golay (SG) smoothing-logistic regression (LR) models achieved 93.33% accuracy of the test set in spectral classification. Moreover, spectral pre-processing and characteristic wavelength selection led to high prediction accuracies for the content of significant components in L. aggregata using standard normal variate (SNV)-competitive adaptive reweighted sampling (CARS)-least squares support vector machine (LSSVM) and SNV-CARS-extreme learning machine (ELM) ( > 0.87 for the test set). This is the first study to provide a visual representation of the content of marker compounds in L. aggregata roots, offering a rapid, non-destructive method for assessing the quality of Linderae Radix. It scientifically justifies the medicinal use of tuberous roots and illuminates rapid quality evaluation through morphological identification. • A rapid, non-destructive method for assessing the quality of Linderae Radix. • Integrated metabolomics, HSI, and machine learning for rapid quality assessment. • Machine learning for establishing classification and quantitative prediction models . • Visualize the spatial distribution of the norisoboldine content in drug slices.