Mislabeling identification of fresh retail beef cuts using machine learning - guided REIMS lipidomic fingerprints

Fraudulent activities such as mislabeling of beef occurred frequently. In this study, a fast and reliable methodology for specific discrimination of fresh retail beef cuts (ribeye, striploin, brisket, beef shank, and fore shank) was established based on electrosurgical knife - rapid evaporative ionization mass spectrometry (iKnife-REIMS) technique integrated with machine learning (ML). REIMS lipidomic fingerprints of various beef cuts were characterized by ML models to explore meaningful information. Authenticity identification models were built by main ML algorithms including discriminant analysis (DA), support vector machines (SVM), and k-nearest neighbor (KNN) to realize the discrimination of mislabeling in beef cuts. Mislabeling discrimination rates of DA, SVM, and KNN were 95.54%, 99.91%, and 100%, respectively. Model performances were evaluated by confusion matrix, receiver operating characteristic curve, and area under the curve. The validation of the proposed REIMS method for detecting mislabeled beef cuts with an accuracy of 96.80% was performed. Results demonstrated that this artificial intelligent method coupled to ML - guided REIMS analysis was efficient for authenticity detection of mislabeled beef cuts.