核磁共振代谢组学数据预处理时常常需要对谱峰面积进行归一化，而归一化后数据的相对比例与原始谱图中峰面积的相对比例相同，因此归一化后数据能反映谱峰面积的相对信息。选取每个样本对应的谱峰面积归一化后数据进行分析，由于成分数据是仅含有相对信息的非负向量，故归一化后的数据可以被考虑为是成分数据。基于核磁共振代谢组学的成分数据分析可研究中药评价中各组样本的均一性、寻找影响不同组分类的特征代谢物、对给定的新样本进行判别分析。黄芪质量评价的实例分析结果证明提出的方法是可行的。

The peak area normalization is often needed as the preprocessing of NMR metabolomic data and the relative ratio of normalized data is the same as that of peak area in original spectrum, so the normalized data can reflect the relative information of peak area. This research selects the normalized data of the peak area for each sample to analyze, compositional data is the nonnegative vector containing only the relative information. Therefore the normalized data can be considered as compositional data. This paper used compositional data analysis based on NMR metabolomics to study the homogeneity for each group's samples, identify the characteristic metabolites which contribute the classification of different groups, and make the discriminate analysis for the given new sample in the evaluation of Chinese materia medica. In the case the analysis of quality evaluation on Astragali Radix, it can be seen from the results that the proposed method is feasible.

山西省高等学校教学改革项目（J2014006）；山西省自然科学基金资助项目（2015011044）；山西省国际交流合作项目（2015081020）