Differentiation of Citri Reticulatae Pericarpium varieties via HPLC fingerprinting of polysaccharides combined with machine learning.

To accurately and reliably distinguish different varieties of Citri Reticulatae Pericarpium (CRP), we propose a novel classification strategy combining polysaccharide fingerprinting and machine learning (ML). First, extraction conditions are optimized using the one-variable-at-a-time method and response surface methodology, and the extraction yield of total polysaccharides reaches 25.15%, with different varieties exhibiting different anti-oxidant abilities. Next, the hydrolysis conditions are optimized for constructing a polysaccharide HPLC fingerprinting, followed by the identification 10 common peaks, including D-Man, L-Rha and D-GalA. Thereafter, among nine supervised ML models, five models with high accuracy (> 0.911) and precision (> 0.926) are selected. Finally, upon combining ML for the classification of CRPs, D-Man, D-Gal, D-Xyl and L-Ara are screened as Q-markers with accuracy, and precision more than 0.944. In summary, we demonstrate the reliability of combining polysaccharide fingerprinting and ML for classifying varieties of CRPs, providing a novel quality evaluation method for the distinguishing natural herbal medicines. CHEMICAL COMPOUNDS STUDIED IN THIS ARTICLE: D-Glucose (PubChem CID: 5793); D-Mannose (PubChem CID: 18950); D-Galactose (PubChem CID: 6036); D-Galacturonic acid (PubChem CID: 439215); D-Xylose (PubChem CID: 135191); L-Rhamnose (PubChem CID: 25310); L-Arabinose (PubChem CID: 439195); Sulphuric acid (PubChem CID: 1118).