Machine Learning for Predicting Zearalenone Contamination Levels in Pet Food

• Published in: Toxins Vol. 16; no. 12; p. 553
• Main Authors: Wang, Zhenlong, An, Wei, Wang, Jiaxue, Tao, Hui, Wang, Xiumin, Han, Bing, Wang, Jinquan
• Format: Journal Article
• Language: English
• Published: Switzerland MDPI AG 01.12.2024; MDPI
• Subjects: Accuracy; Acute toxicity; Algorithms; Animal Feed - analysis; Animal Feed - microbiology; Animals; Classification; Communication; Contamination; Cost effectiveness; Data mining; E-nose; Electronic Nose; Electronic noses; Food; Food contamination; Food contamination & poisoning; Food Contamination - analysis; Food law; Health problems; Learning algorithms; Legislation; Liquid chromatography; Machine Learning; Mass spectrometry; Mass spectroscopy; Metal oxides; Methane; Multilayer perceptrons; Multilayers; mycotoxin; Mycotoxins; Performance evaluation; Pet food; Pet supplies industry; Petfoods; Pets; Positron emission; Safety and security measures; Sensors; Sulfur; Support vector machines; Tandem Mass Spectrometry; VOCs; Volatile compounds; Volatile organic compounds; Volatile Organic Compounds - analysis; Zearalenone; Zearalenone - analysis; China; mycotoxin; pet food; zearalenone; machine learning; E-nose
• ISSN: 2072-6651; 2072-6651
• DOI: 10.3390/toxins16120553

Zearalenone (ZEN) has been detected in both pet food ingredients and final products, causing acute toxicity and chronic health problems in pets. Therefore, the early detection of mycotoxin contamination in pet food is crucial for ensuring the safety and well-being of animals. This study aims to develop a rapid and cost-effective method using an electronic nose (E-nose) and machine learning algorithms to predict whether ZEN levels in pet food exceed the regulatory limits (250 µg/kg), as set by Chinese pet food legislation. A total of 142 pet food samples from various brands, collected between 2021 and 2023, were analyzed for ZEN contamination via liquid chromatography–tandem mass spectrometry. Additionally, the “AIR PEN 3” E-nose, equipped with 10 metal oxide sensors, was employed to identify volatile compounds in the pet food samples, categorized into 10 different groups. Machine learning algorithms, including liner regression, k-nearest neighbors, support vector machines, random forests, XGBoost, and multi-layer perceptron (MLP), were used to classify the samples based on their volatile profiles. The MLP algorithm showed the highest discrimination accuracy at 86.6% in differentiating between pet food samples above and below the ZEN threshold. Other algorithms showed moderate accuracy, ranging from 77.1% to 84.8%. The ensemble model, which combined the predictions from all classifiers, further improved the classification performance, achieving the highest accuracy at 90.1%. These results suggest that the combination of E-nose technology and machine learning provides a rapid, cost-effective approach for screening ZEN contamination in pet food at the market entry stage.