Bouhlel Jihéne, Jouan-Rimbaud Bouveresse Delphine, Abouelkaram Said, Baéza Elisabeth, Jondreville Catherine, Travel Angélique, Ratel Jérémy, Engel Erwan, Rutledge Douglas N
UMR Ingénierie Procédés Aliments, AgroParisTech, Inra, Université Paris-Saclay, 91300 Massy, France; INRA, UR 370 QuaPA, MASS Laboratory, Saint-Genès-Champanelle, France.
UMR Ingénierie Procédés Aliments, AgroParisTech, Inra, Université Paris-Saclay, 91300 Massy, France.
Talanta. 2018 Feb 1;178:854-863. doi: 10.1016/j.talanta.2017.10.025. Epub 2017 Oct 18.
The aim of this work is to compare a novel exploratory chemometrics method, Common Components Analysis (CCA), with Principal Components Analysis (PCA) and Independent Components Analysis (ICA). CCA consists in adapting the multi-block statistical method known as Common Components and Specific Weights Analysis (CCSWA or ComDim) by applying it to a single data matrix, with one variable per block. As an application, the three methods were applied to SPME-GC-MS volatolomic signatures of livers in an attempt to reveal volatile organic compounds (VOCs) markers of chicken exposure to different types of micropollutants. An application of CCA to the initial SPME-GC-MS data revealed a drift in the sample Scores along CC2, as a function of injection order, probably resulting from time-related evolution in the instrument. This drift was eliminated by orthogonalization of the data set with respect to CC2, and the resulting data are used as the orthogonalized data input into each of the three methods. Since the first step in CCA is to norm-scale all the variables, preliminary data scaling has no effect on the results, so that CCA was applied only to orthogonalized SPME-GC-MS data, while, PCA and ICA were applied to the "orthogonalized", "orthogonalized and Pareto-scaled", and "orthogonalized and autoscaled" data. The comparison showed that PCA results were highly dependent on the scaling of variables, contrary to ICA where the data scaling did not have a strong influence. Nevertheless, for both PCA and ICA the clearest separations of exposed groups were obtained after autoscaling of variables. The main part of this work was to compare the CCA results using the orthogonalized data with those obtained with PCA and ICA applied to orthogonalized and autoscaled variables. The clearest separations of exposed chicken groups were obtained by CCA. CCA Loadings also clearly identified the variables contributing most to the Common Components giving separations. The PCA Loadings did not highlight the most influencing variables for each separation, whereas the ICA Loadings highlighted the same variables as did CCA. This study shows the potential of CCA for the extraction of pertinent information from a data matrix, using a procedure based on an original optimisation criterion, to produce results that are complementary, and in some cases may be superior, to those of PCA and ICA.
本研究旨在将一种新型探索性化学计量学方法——共同成分分析（CCA），与主成分分析（PCA）和独立成分分析（ICA）进行比较。CCA是通过将称为共同成分和特定权重分析（CCSWA或ComDim）的多块统计方法应用于单个数据矩阵（每个块一个变量）而形成的。作为应用，将这三种方法应用于肝脏的固相微萃取-气相色谱-质谱挥发物组学特征，以试图揭示鸡暴露于不同类型微污染物的挥发性有机化合物（VOC）标志物。将CCA应用于初始的固相微萃取-气相色谱-质谱数据时，发现样本得分沿CC2存在漂移，这是进样顺序的函数，可能是由于仪器中与时间相关的变化所致。通过对数据集相对于CC2进行正交化消除了这种漂移，所得数据用作输入到这三种方法中的正交化数据。由于CCA的第一步是对所有变量进行归一化缩放，因此初步数据缩放对结果没有影响，所以CCA仅应用于正交化的固相微萃取-气相色谱-质谱数据，而PCA和ICA分别应用于“正交化”、“正交化和帕累托缩放”以及“正交化和自动缩放”的数据。比较结果表明，PCA结果高度依赖于变量的缩放，而ICA则不同，数据缩放对其影响不大。然而，对于PCA和ICA，在对变量进行自动缩放后，暴露组的分离最为明显。本研究的主要部分是将使用正交化数据的CCA结果与应用于正交化和自动缩放变量的PCA和ICA结果进行比较。通过CCA获得了暴露鸡组最清晰的分离。CCA载荷也清楚地识别出对共同成分贡献最大从而导致分离的变量。PCA载荷没有突出每次分离中最具影响力的变量，而ICA载荷突出的变量与CCA相同。这项研究表明，CCA有潜力从数据矩阵中提取相关信息，它使用基于原始优化标准的程序，产生的结果具有互补性，在某些情况下可能优于PCA和ICA。
BMC Bioinformatics. 2010-4-15
Hum Brain Mapp. 2021-6-1
Zotero 插件