Machine learning-guided Orbitrap-HRAMS-based metabolomic fingerprinting for geographical origin, variety and tissue specific authentication, and adulteration detection of turmeric and ashwagandha

Abstract

The increasing global demand for herbs and spices in food and nutraceutical industries highlights their key functional benefits, including antioxidant and anti-inflammatory properties. Ensuring authenticity and traceability is essential to counteract challenges such as geographical origin (GO) mislabelling and tissue- or variety-specific adulteration, which can undermine product quality and safety. This study employs LC-Orbitrap-MS-based untargeted metabolomics coupled with machine learning to authenticate the GO, variety, and tissue specificity of turmeric (Curcuma longa) and ashwagandha (Withania somnifera), two widely used food ingredients. Four GO-specific turmeric samples, three tissue- and variety- specific ashwagandha samples, and adulterated market samples were analysed using data-dependent acquisition mode. Machine learning algorithms identified key biomarkers and constructed robust classification models, achieving 98 % specificity and accuracy in authenticating GO, variety, and tissue specificity, even in adulterated samples. These results demonstrate the value of integrating advanced metabolomics and machine learning for quality assurance and food safety in the global market.