Machine learning-enabled flavoromics decoding the molecular basis of aroma profiles in jasmine essential oil under different extraction processes.

Jasmine essential oil holds significant importance in fragrance and food industries due to its unique aroma and bioactivity. However, traditional extraction processes suffer from low yields and severe flavor losses. This study employed flavoromics and machine learning to analyze aroma quality of oils obtained through four different extraction methods, including supercritical CO₂ extraction (SFE-CO₂) and ultrasonic-assisted hydrodistillation (UAHD). Results demonstrated that SFE-CO₂ significantly improved yield and antioxidant activity, while UAHD exhibited highest ACE inhibitory activity. SFE-CO₂ possessed distinctive floral and sweet aromas, whereas UAHD demonstrated prominent fruity aroma. GC-IMS and GC-MS identified 121 volatile components, including 65 characteristic volatiles and 57 quantitatively verified key aroma compounds. XGBoost-SHAP machine learning modeling revealed that citronellol and α-Ionone exhibited synergistic effects on floral aroma formation, hexyl acetate showed inhibitory effects on fatty odor, and limonene's masking effect on grassy odor improved overall aroma. This study provides theoretical foundations for industrial jasmine oil production.