Insight from untargeted metabolomics: Revealing the potential marker compounds changes in refrigerated pork based on random forests machine learning algorithm

Abstract

Data on changes in non-volatile components and metabolic pathways during pork storage were inadequately investigated. Herein, an untargeted metabolomics coupled with random forests machine learning algorithm was proposed to identify the potential marker compounds and their effects on non-volatile production during pork storage by ultra-high-performance liquid chromatography-mass spectrometry (UHPLC-MS/MS). A total of 873 differential metabolites were identified based on analysis of variance (ANOVA). Bioinformatics analysis shows that the key metabolic pathways for protein degradation and amino acid transport are amino acid metabolism and nucleotide metabolism. Finally, 40 potential marker compounds were screened using the random forest regression model, innovatively proposing the key role of pentose-related metabolism in pork spoilage. Multiple linear regression analysis revealed that d-xylose, xanthine, and pyruvaldehyde could be key marker compounds related to the freshness of refrigerated pork. Therefore, this study could provide new ideas for the identification of marker compounds in refrigerated pork.

Keywords: Freshness; Machine learning; Marker compounds; Metabolomics; Random forests; Refrigerated pork.