Integrating electronic sensory, untargeted metabolomics, and FT-NIR spectroscopy with machine learning for rapid origin discrimination and quality assessment of Chebulae Fructus

The accurate botanical origin identification of Chebulae Fructus (CF) is crucial for ensuring their quality and therapeutic efficacy. To address this, this study established, for the first time, a comprehensive quality assessment strategy integrating multiple modern analytical techniques. Initially, DNA barcoding was employed to verify the botanical source of the samples. Subsequently, electronic eyes (E-eye) and electronic noses (E-nose) were employed to objectively quantify and differentiate the colour characteristics and flavour profiles of the fruit. Untargeted metabolomic analysis identified 9037 metabolites, comprehensively revealing distinct chemical profiles between the two subspecies. Antioxidant activity assays indicate that Terminalia chebula Retz. demonstrates superior performance in both the ABTS and FRAP tests. To achieve rapid quality evaluation, machine learning models based on fourier transform near-infrared (FT-NIR) spectroscopy were developed. A classification model integrating linear discriminant analysis (LDA) achieved perfect discrimination between the two origins. Furthermore, partial least squares (PLS) regression models constructed using optimised spectral variables selected via algorithms such as competitive adaptive reweighted sampling (CARS) can accurately predict the content of five key active components and three antioxidant activity indices (R 2 c, R 2 p > 0.92). This approach provides an efficient and reliable analytical tool for the quality control of CF, meeting the industry's demand for rapid, on-site analysis.