南湖新闻网讯（通讯员 肖仕杰）近日，我校海燕策略研究线路一王巧华教授团队研究成果以“Rapid identification of A1 and A2 milk based on the combination of mid-infrared spectroscopy and chemometrics”为题在Food Control发表。研究揭示了A1和A2牛奶的光谱差异，建立了无损检测两种奶的分类模型，表明中红外光谱技术可作为无损快速鉴别A1奶和A2奶的新工具。该研究也可为单独组建A1型和A2型奶牛育种群提供相应的技术支持。
仅含A2β-酪蛋白的牛奶（A2奶）因其独特的健康益处而在全球广受欢迎。长期以来，企业需要先对奶牛进行专业的基因检测，筛选出β-酪蛋白中只包含A2β-酪蛋白的纯种A2奶牛，再用这些奶牛生产的牛乳加工成A2奶。基因检测虽准确性高，但成本也高且耗时长，无法满足乳企规模化生产的要求。因此，急需开发一种低成本、高效益的技术快速识别A1奶（普通奶）和A2奶。该研究突破了传统基因检测的局限，应用中红外光谱技术快速鉴别出A1和A2牛奶。
CARS算法筛选特征变量
该研究分析A1和A2牛奶在中红外光谱吸光度上的差异，找到敏感波段组合作为全光谱，分别利用标准正态变量变换 、多元散射校正、归一化、一阶导数、二阶导数、一阶差分和二阶差分等7种方法对光谱进行预处理，利用无信息变量消除法和竞争性自适应重加权算法筛选出能代表A1和A2奶差异的特征变量，进而构建偏最小二乘判别分析（PLS-DA）模型和支持向量机(SVM)模型，PLS-DA模型的训练集准确率和测试集准确率分别为96.6%和96.0%，SVM模型的训练集准确率和测试集准确率分别为96.0%和95.1%。
UVE算法筛选特征变量
该研究选择PLS-DA模型作为最佳模型，使用一组独立样本对模型进行外部验证。将新采集的牛奶中红外光谱批量带入保存的模型中，以对应的奶牛基因检测结果作为对照指标，模型的预测准确率为95.2%，性能良好。结果表明，中红外光谱技术可以实现对A1奶与A2奶的快速分类鉴别，有望将来在生产中得到应用。
海燕策略研究线路一硕士研究生肖仕杰为论文第一作者，海燕策略研究线路一王巧华教授和动物科学技术学院张淑君教授为共同通讯作者。该研究得到中国政府项目（2013070204020045）资助。
审核人：王巧华
【英文摘要】
The milk containing only A2 β-casein (called A2 milk) is globally popular because of its unique health benefits. Traditionally, genetic testing (such as gene sequencing) is used to identify the cows with A2 β-casein gene that can only produce A2 milk, which is a time-consuming and costly method. The objective of this study was to directly identify A1 and A2 milk from a large quantity of milk using mid-infrared (MIR) spectroscopy and chemometrics without genotyping cows. Before establishing the predictive model, we firstly genotyped the A1 β-casein and A2 β-casein of cows from blood as reference values. Further, the MIR spectra of the milk collected from these cows were obtained using a dairy product analyzer. The MIR spectroscopy data and the reference values were used as the independent and dependent variables, respectively, to establish a category classification model for A1 and A2 milk. Seven preprocessing methods were combined with two feature extraction algorithms to establish the model. Subsequently, partial least squares discriminant analysis (PLS-DA) and support vector machine (SVM) models were developed. The average accuracy of the test set of the two models were 94.9% and 94.4%, respectively, while the PLS-DA model exhibited better effect, and the accuracy of training set and test set reached 96.6% and 96.0%, respectively. We used a set of independent samples for the external validation of the PLS-DA model, and the prediction accuracy was 95.2%. Overall, the proposed prediction models based on MIR spectroscopy can be used for low-cost, rapid, and large-scale classification of A1 and A2 milk, which may be extremely beneficial in milk production industries.
论文链接：https://doi.org/10.1016/j.foodcont.2021.108659
