Novel method for rapid identification of Listeria monocytogenes based on metabolomics and deep learning. (September 2022)

• Record Type: Journal Article
• Authors: Feng, Ying; Cheng, Zhangkai J.; Wei, Xianhu; Chen, Moutong; Zhang, Jumei; Zhang, Youxiong; Xue, Liang; Chen, Minling; Li, Fan; Shang, Yuting; Liang, Tingting; Ding, Yu; Wu, Qingping
• Is Part Of: Food control. Volume 139(2022)
• Publication Date: 2022-09
• Subjects: Metabolomics -- Listeria monocytogenes -- Pathogen identification -- Deep learning; Food -- Quality -- Periodicals; Food -- Analysis -- Periodicals; Food handling -- Periodicals; Food industry and trade -- Quality control -- Periodicals; Aliments -- Industrie et commerce -- Qualité -- Contrôle -- Périodiques; Aliments -- Qualité -- Périodiques; Aliments -- Analyse -- Périodiques; Hygiène alimentaire -- Périodiques; Food -- Analysis; Food handling; Food -- Quality; Periodicals; Electronic journals; 664.07
• DOI: 10.1016/j.foodcont.2022.109042
• Languages: English
• ISSNs: 0956-7135

Abstract

Abstract: Metabolomics based on the mass spectrometry approach can serve as a platform to detect pathogens and spoilage microorganisms. However, the accurate quantification of biomarkers with lower molecular weight based on mass spectrometry is generally limited by isotope-labeled standards and complicated protocols, which is not conducive to large-scale applications. Here, we developed a novel method that combined metabolomics with deep learning for the identification of Listeria monocytogenes . A convolutional neural network (CNN) model of these three potential biomarkers for L. monocytogenes was established, with a prediction accuracy of 82.2%. Furthermore, metabolic fingerprints composed of 29 metabolites were obtained using pseudotargeted metabolomics approach, which successfully distinguished six common Listeria species in hierarchical cluster analysis. The binary and multiple classifiers of CNN models were established to identify L. monocytogenes and common pathogens, which prediction accuracies were 96.7% and 96.3%, respectively. This novel method combined pseudotargeted metabolomics with deep learning is a promising powerful tool for pathogen identification and classification. Highlights: A platform that combined metabolomics and deep learning was used for pathogen identification. Three metabolic potential biomarkers for identifying L. monocytogenes were selected. The CNN model based on metabolic fingerprints was established. Compared to traditional LC–MS methods, the detection time is shortened to 12 min.