Integrated HRMS metabolomics and machine learning strategy for discovering discriminant markers of Fritillaria and authenticity verification in commercial products.

The authenticity of Fritillaria is of critical importance given the widespread use of these bulbs in traditional medicine and commercial products. The prized Fritillaria cirrhosa is frequently adulterated with cheaper look-alike species. To address this, an integrated analytical strategy combining HRMS metabolomics with complementary machine learning algorithms was developed to discover chemical markers for species discrimination. OPLS-DA initially screened 60 candidate compounds from authenticated Fritillaria cirrhosa and Fritillaria ussuriensis bulbs. Subsequent feature selection using random forest and support vector machine yielded a balanced 10-marker panel capturing the unique metabolic signatures of both species. Six of ten commercial Fritillaria cirrhosa were unambiguously identified as Fritillaria ussuriensis, revealing a high risk of adulteration. Nine markers were tentatively identified, and one diagnostic MS/MS fragment was discovered to differentiate the epimeric alkaloids peiminine and sipeimine. This study provides a robust tool for quality control of Fritillaria and demonstrates the advantage of combining multiple machine learning models for unbiased marker selection.