Abstract
The in-situ, rapid and non-destructive identification of complex food samples is a long-standing core technical challenge in food chemistry. Herein, vapor assisted desorption chemical ionization mass spectrometry (VADCI-MS) device was independently constructed, which facilitates rapid and non-destructive analysis of complex foods with diverse morphologies. Additionally, an integrated strategy that combines MS fingerprinting, feature selection, and machine learning classification models was established and is particularly effective for complex foods that are challenging to differentiate using conventional methods. As a paradigm, VADCI-MS was employed to rapidly (<1.0 min/sample) and non-destructively obtain the MS fingerprints of 101 batches of Zanthoxylum bungeanum (ZB) samples and 415 feature peaks were selected. Further, using machine learning classification models, a high accuracy differentiation of ZB from different origins was achieved (accuracy rate = 96.88%) and was verified to be substantially improved over traditional GC-MS. Conclusively, VADCI-MS coupled with machine learning holds significant potential for accurate and non-destructive identification of foods.