Volatile Fingerprinting and Interpretable Machine Learning for Quality Differentiation of Astragali Radix from...

Area of Science:

• Analytical Chemistry
• Chemometrics
• Natural Product Chemistry

Background:

• Astragali Radix (AR) is a valuable food-medicine plant, but its volatile profiles differ based on cultivation.
• Understanding these differences is crucial for authentication and quality control of AR.

Purpose of the Study:

• To characterize and differentiate volatile fingerprints of wild, wild-simulated, and cultivated Astragali Radix (AR).
• To identify key volatile compounds that distinguish AR samples based on their cultivation patterns using integrated analytical techniques and machine learning.

Main Methods:

• Integrated headspace solid-phase microextraction-gas chromatography-mass spectrometry (HS-SPME-GC-MS) and headspace gas chromatography-ion mobility spectrometry (HS-GC-IMS) for volatile profiling.
• Applied multivariate analysis (PLS-DA) and interpretable machine learning (Random Forest, SVM, CatBoost) with VIP and SHAP values for compound identification.
• Analyzed 117 AR samples from diverse cultivation origins.

Main Results:

• Distinct volatile fingerprints were observed across wild, wild-simulated, and cultivated AR samples.
• HS-GC-IMS provided clearer group separation compared to HS-SPME-GC-MS.
• Identified candidate discriminatory compounds including esters, aldehydes, alcohols, and ketones, potentially linked to fatty acid, aromatic, and terpenoid metabolism.

Conclusions:

• The combined analytical approach effectively captured cultivation-pattern-associated volatile differences in AR.
• Identified specific volatile compounds (e.g., γ-hexalactone, methyl eugenol, 1-octen-3-one) can serve as markers for AR authentication.
• The developed workflow offers a robust strategy for rapid screening and quality assessment of AR based on volatile profiles.