Deep Eutectic Solvent Extraction of Heteratisine from Aconitum tanguticum: COSMO-RS Screening, RSM Optimization and Machine Learning Comparison

Aconitum tanguticum (Maxim.) Stapf is a Tibetan medicinal plant with low toxicity. To obtain the major alkaloid heteratisine (HTS) for bioactivity evaluation, this study established a method combining deep eutectic solvent (DES) with ultrasound-assisted extraction. COSMO-RS screening identified menthol-octanoic acid (1:2) as the optimal solvent, giving a content of 0.321 mg/g, approximately 9.17 times that of ethanol. Extraction optimization was performed via response surface methodology on a 29-run Box–Behnken design, providing a validated optimum of 0.499 mg/g at 256.8 W, 30.5 min, 1:30.3 g/mL, and 39.7 °C. The fitting behavior of three machine learning models and quadratic polynomial regression (QPR) was compared on a 75-run non-structured dataset. XGBoost and GPR achieved high in-sample fitting performance (R2 = 0.983 and 0.994), whereas QPR and Random Forest showed limited fitting (0.529 and 0.546). Condition-grouped nested validation demonstrated that none of the models could extrapolate beyond the observed dataset. Thus, the ML analyses were framed as in-sample diagnostic tools, with GPR offering uncertainty estimates to identify data-sparse regions. In HEK293T-TRPV1 cells, HTS pretreatment reduced capsaicin-induced Ca2+ signaling. In general, this study provides a DES-based extraction approach and a multi-model comparison framework for the utilization of A. tanguticum.