Mannu Alberto, Karabagias Ioannis K, Di Pietro Maria Enrica, Baldino Salvatore, Karabagias Vassilios K, Badeka Anastasia V
Department of Chemistry, University of Turin, Via Pietro Giuria, 7, I-10125 Turin, Italy.
Laboratory of Food Chemistry, Department of Chemistry, University of Ioannina, 45110 Ioannina, Greece.
Foods. 2020 Aug 2;9(8):1040. doi: 10.3390/foods9081040.
A fast, economic, and eco-friendly methodology for the wine variety and geographical origin differentiation using C nuclear magnetic resonance (NMR) data in combination with machine learning was developed. Wine samples of different grape varieties cultivated in different regions in Greece were subjected to C NMR analysis. The relative integrals of the C spectral window were processed and extracted to build a chemical fingerprint for the characterization of each specific wine variety, and then subjected to factor analysis, multivariate analysis of variance, and -nearest neighbors analysis. The statistical analysis results showed that the C NMR fingerprint could be used as a rapid and accurate indicator of the wine variety differentiation. An almost perfect classification rate based on training (99.8%) and holdout methods (99.9%) was obtained. Results were further tested on the basis of Cronbach's alpha reliability analysis, where a very low random error (0.30) was estimated, indicating the accuracy and strength of the aforementioned methodology for the discrimination of the wine variety. The obtained data were grouped according to the geographical origin of wine samples and further subjected to principal component analysis (PCA) and partial least squares-discriminant analysis (PLS-DA). The PLS-DA and variable importance in projection (VIP) allowed the determination of a chemical fingerprint characteristic of each geographical group. The statistical analysis revealed the possibility of acquiring useful information on wines, by simply processing the C NMR raw data, without the need to determine any specific metabolomic profile. In total, the obtained fingerprint can be used for the development of rapid quality-control methodologies concerning wine.
开发了一种快速、经济且环保的方法，用于结合机器学习利用碳核磁共振（NMR）数据对葡萄酒品种和地理来源进行区分。对希腊不同地区种植的不同葡萄品种的葡萄酒样品进行了碳NMR分析。对碳光谱窗口的相对积分进行处理和提取，以构建用于表征每个特定葡萄酒品种的化学指纹图谱，然后进行因子分析、多变量方差分析和K近邻分析。统计分析结果表明，碳NMR指纹图谱可作为葡萄酒品种区分的快速准确指标。基于训练（99.8%）和留出法（99.9%）获得了几乎完美的分类率。根据克朗巴哈系数可靠性分析进一步测试结果，其中估计随机误差非常低（0.30），表明上述葡萄酒品种鉴别方法的准确性和可靠性。根据葡萄酒样品的地理来源对获得的数据进行分组，并进一步进行主成分分析（PCA）和偏最小二乘判别分析（PLS-DA）。PLS-DA和投影变量重要性（VIP）使得能够确定每个地理组的化学指纹特征。统计分析表明，通过简单处理碳NMR原始数据，无需确定任何特定的代谢组学谱，就有可能获得有关葡萄酒的有用信息。总体而言，所获得的指纹图谱可用于开发有关葡萄酒的快速质量控制方法。
