Discrimination of thermal treated bovine milk using MALDI-TOF MS coupled with machine learning

Thermal sterilization influenced the molecular composition and quality of milk and dairy products. Distinguishing thermal treated milk samples under different conditions has the practical significance for the authentication and quality control of milk products. In this study, a simple and accurate method for the discrimination of thermal processed milk samples based on matrix assisted laser desorption time-of-flight mass spectrometry (MALDI-TOF MS) and machine learning was established. Milk samples collected from local dairy farms were incubated with varying temperature and heating duration to cover a wide range of heat load. MS signals of featured peptides were selected by three orthogonal statistical methods, including partial least-squares-discriminant (PLS-DA), least absolute shrinkage and selection operator (LASSO) and recursive feature elimination with cross-validation (RFECV). The featured peptides were used as training and testing variables for machine learning. A total of 14 algorithms were implemented and evaluated to construct machine learning model to discriminate thermal treated milk samples. Finally, the model was evaluated and compared by accuracy, recall, F1 value and accuracy, and the optimal model was selected. The results showed support vector machine with a linear kernel (SVM-L), random forest (RF), linear discriminant analysis (LDA), and penalized discriminant analysis (PDA) have the best performance, with prediction accuracies of 0.97, 0.96, 0.96, and 0.96, respectively. These results demonstrate that machine learning algorithm can be applied to identify thermal treated milk samples by combining MALDI-TOF MS analysis.