The aim of this work was to develop an approach combining LC-MS-based metabolomics and machine learning to distinguish between and predict authentic and adulterated lemon juices. A targeted screening of six major flavonoids was first conducted using ultraviolet ion trap MS. To improve the prediction accuracy, an untargeted methodology was carried out using UHPLC-QTOF/MS. Based on the acquired metabolic profiles, both PCA and PLS-DA were conducted. Results exhibited a cluster pattern and a separation potential between authentic and adulterated samples. Five machine learning models were then developed to further analyze the data. The model of support vector machine achieved the highest prediction power, with accuracy up to 96.7 ± 7.5% for the cross-validation set and 100% for the testing set. In addition, 79 characteristic m/z were tentatively identified. This work demonstrated that untargeted screening coupled with machine learning models can be a powerful tool to facilitate detection of lemon juice adulteration.

这项工作的目的是开发一种将基于 LC-MS 的代谢组学和机器学习相结合的方法，以区分和预测真正的和掺假的柠檬汁。首先使用紫外线离子阱 MS 对六种主要黄酮类化合物进行了有针对性的筛选。为了提高预测准确性，使用 UHPLC-QTOF/MS 进行了非靶向方法。基于获得的代谢谱，进行了 PCA 和 PLS-DA。结果显示了真实样本和掺假样本之间的聚类模式和分离潜力。然后开发了五个机器学习模型来进一步分析数据。支持向量机模型实现了最高的预测能力，交叉验证集的准确率高达 96.7±7.5%，测试集的准确率高达 100%。此外，79 特征m/ z进行了初步鉴定。这项工作表明，无目标筛选与机器学习模型相结合可以成为促进柠檬汁掺假检测的强大工具。