Machine learning-driven assessment of origin quality for the primary active ingredient in Gastrodia elata: towards precision cultivation

Abstract

Gastrodia elata is a traditional and valuable Chinese medicinal herb that possesses both medicinal and edible properties; its clinical efficacy is significantly influenced by its quality. However, as cultivation expands, the quality of G. elata from different producing areas exhibits considerable variability. Therefore, a systematic evaluation of G. elata from different origins is essential for cultivating high-quality medicinal materials. This study innovatively integrates multi-source active ingredient data with environmental factors and employs a machine learning-SHAP analysis framework to explore the above scientific issues, thereby achieving precise quantification of factor contributions and spatial prediction of quality. The results indicate that the quality of wild G. elata surpasses that of cultivated varieties, with significantly higher levels of gastrodin, 4-Hydroxybenzyl alcohol, parishin A, and parishin B. Notably, 96% of cultivated G. elata meet the basic requirements outlined in the Chinese Pharmacopoeia, although the contents of effective components vary significantly across regions. Among these, 4-Hydroxybenzyl alcohol emerges as the most significant effective component for distinguishing G. elata from different origins, followed closely by gastrodin. Spatial distribution maps reveal that regions with high concentrations of gastrodin and 4-Hydroxybenzyl alcohol are predominantly located in the traditionally interconnected areas of eastern Sichuan, northeastern Yunnan, and western Guizhou. Furthermore, SHAP analysis identifies the mean temperature of the coldest quarter as the most critical determinant affecting its quality. This study not only confirms traditional knowledge but also uncovers new patterns, thereby providing important scientific evidence and data support for the standardized cultivation and systematic quality evaluation of G. elata.