基于化学计量学的特种家畜乳及乳制品真实性鉴别模型建立与优化

Traditional food authentication or discrimination by single marker or a few markers belongs to the exclusion approaches, only can be used as marker to reject and disapprove. While Chemometrics based big data structure analysis, data visualization, pattern analysis and discriminant analysis make it possible to profiling or make portrait of food properties from different aspects, and this implies providing brand new concept and strategies to the more and more sophisticated food authentication needs and challenges. With the rapid development of cow and goat milk industry in recent years,peculiar livestock milk, such as horse, camel, yak and buffalo milk are also more and more popularizing in China. Because of its rareness, special flavor, high nutritive value, and some health function, the prices of these peculiar milk are rocketing, several, even more than ten times higher over cow or goat milk, thus very vulnerable to fraud and adulteration with cow or goad milk, or by food industrial materials (such as non-dairy milk powder, dextrin, and vegetable protein et al). Milk fraud recognition/discrimination by gene and species specific protein based methods only workable to species simulate, and easily compromised by heating, homogenizing and fermentation et al food processing treatment.. Horse, camel, cow and goat milk of Inner Mongolia, as well as yak and buffalo milk in other province, will be systematically gathered as research sample sets, and certain sets of nutrients fingerprint (such as fatty acids, triglycerides, amino acids and vitamins profiles；i.e. data arrays) and analytical instrument fingerprints(such as chromatographic, mass, and near infrared spectrum profiles) will be derived as well. Appropriate Chemometrics algorithm and pattern and discrimination analysis modules(such as PCA, PLS, SIMCA,and PLS-DA) and artificial intelligence modules (such as SVM, ANN) will be selected and combined in the fingerprint data structure analysis, building and optimizing the animals milk and product authentication models, to establish a set of robust, universal and rapid detection approaches and authentication model.

传统单指标或少数指标的孤立分析，属排除法策略，通常可否定而不能肯定。化学计量学大数据结构分析和可视化、模式和判别分析，使正面描绘食物特征成为可能，为纷繁复杂的食物真实性鉴别需求和挑战提供了全新思路和手段。我国牛羊乳业发展的同时，马、骆驼、牦牛和水牛等家畜乳也得到了区域性开发。由于其稀缺、风味独特、营养价值高、又具保健功能而供不应求，价格高于牛羊奶几倍、十几倍，极易被低价位牛羊奶或其他原料（如植脂末等等）掺假或冒充。基于蛋白和DNA的物种鉴别，易受加热、均质和发酵等加工的干扰，且不能鉴定非物种异常。本项目以我区马和骆驼乳为主要研究对象，兼顾牦牛和水牛乳，系统采集样本集的某些营养素和分析仪器指纹（一组或多组指标集），选择和组合化学计量学演算和判别分析模块，进行指纹数据特征分析，建立和优化各种家畜乳及其制品的判别模型，以期建立稳健的、通用和快速的特种乳真实性判别模型及整套分析方案。

我国乳业进入了多样化、特色化高品质发展时代，特种家畜乳得到区域性开发，因其风味独特、营养价值高且具保健功能，但又产量小，价格高于牛羊奶几倍至十几倍，极易被冒充或掺假。当前食品完整性（Food Integrity）和真实性（Food Authenticity）受到食品监管体系和消费者共同关注，成为国际食品贸易新的市场准入准则和技术壁垒，特种家畜乳指纹特征和真实性鉴别方法学研究十分必要。. 本研究较为系统地研究了六种家畜乳脂肪酸（FAs）、甘油三酯（TAGs）、氨基酸（AAs）、气味物质、近红外（NIR）光谱和质谱图数据指纹，发现均有不同程度的物种和饲养模式特异性，且与复杂的生产加工及品牌和商品定位等“人为”的“工业”因素相关。首次提出了乳制品“工业指纹”概念。六种家畜乳和乳制品本均可以按上述指纹自然聚类，符合或基本符合物种分类学规律。牛科的荷斯坦牛、牦牛、山羊和水牛乳聚类同一个相区，双峰驼和蒙古马乳分别远远聚类于另外两个相区。乳脂质谱、NIR光谱和气味质谱指纹易受多种因素干扰，模型不如营养素的稳健，但可开发快速判别模型，且具较大的优化空间和开发价值。NIR对非乳成分掺假有较强的区分能力，对进口和国产荷斯坦牛乳粉也有很好的分离，值得深入研究。马和驼独特的维生素C和A可纳入判别模型指纹库。. 成功构建了基于上述指纹的物种和成分真实性判别模型，有机和非有机乳判别模型。优化的各模型外部检验准确率均在90%以上，马乳和驼乳有效掺假冒充的判别准确率可达100%，单企业样本建立的模型对本企业有机和非有机乳判别准确率可达100%；NIR线性模型可准确判别乳粉中掺假5%以上的植脂末、糊精、植物分离蛋白粉和乳清粉。. 基于化学计量学食品指纹特征分析具有巨大的分类和判别分析能力，并具有学习性和开放性，为乳及乳制品真实性判别和溯源探索开发带来了无限可能。在矿物质和稳定同位素指纹、草饲和谷饲、民族传统乳制品，以及我区优势特色六畜肉资源均可拓展研究。