Data-driven spectral stitching improves the performance of flow injection-mass spectrometry for food authentication:...

Pomegranate juice is frequently adulterated with cheaper fruit species (e.g., apple, pear, and grape). A rapid flow-injection-mass spectrometry (FI-MS) method coupled with machine learning (ML) was developed for pomegranate juice authentication. A data-driven spectral-stitching strategy was designed to improve authentication accuracy. After comparing different ML algorithms and models constructed using different datasets, an optimized hierarchical classification workflow was established. Specifically, a 7-class random forest (RF) model was used to differentiate 4 species of pure juices and 3 types of adulterated juice, followed by a semi-quantification model to recognize adulteration level. This workflow detected apple and pear at low concentrations (1-5%), and ∼ 10% for grape due to the chemical similarity with pomegranate. The 100% accuracy for recognizing commercial pure pomegranate juice further validated robustness of the workflow. This optimized workflow combining spectral-stitching FI-MS with RF modeling provides a rapid, reproducible framework for juice authentication that is extendable to other food commodities.