质谱代谢组学数据预处理方法研究  

Research on the preprocessing method of mass metabolomics data

作　　者：刘月程 王焕军 马金刚[1] 刘静[1] 魏德健[1] 周晨烁 曹慧[1] LIU Yuecheng;WANG Huanjun;MA Jingang;LIU Jing;WEI Dejian;ZHOU Chenshuo;CAO Hui(Shandong University of Traditional Chinese Medicine,Jinan 250355,China)

机构地区：[1]山东中医药大学,济南250355

出　　处：《化学分析计量》2018年第5期105-109,共5页Chemical Analysis And Meterage

基　　金：国家自然科学基金面上项目(81473653);山东省博士基金(ZR2016HB50);中国博士后科学基金特别资助项目(2015T80741);山东省社会科学规划研究项目(17CHLJ09)

摘　　要：建立质谱代谢组学数据的预处理方法。采用K近邻法、连续K近邻法、多重插补法进行缺失值填补,应用正态、极差、均值、中位数、中心化、总强度标准化等方法进行数据预处理,以正态性检验、模型拟合能力、预测能力、分类效果对预处理方法进行评价。通过模型拟合能力、预测能力、分类效果检验评价,缺失值填补3种方法差别较小;直观分析确定多重插补法,正态性检验结果表明,经过缺失值插补后数据呈偏态分布;极差标准化后,数据呈现正态分布。6种标准化方法的评价结果显示,极差标准化后数学建模的拟合能力、预测能力和分类效果较好。应用多重插补缺失值填补–极差标准化法得到的质谱代谢组学数据可以进行数据模式识别。The preprocessing method of mass metabolomics data was established.The K neighbor method,continuous K neighbor method and multiple interpolation method were employed to fill the missing value.The normalizing,min–max,mean,median,centralization,total strength methods were used for data processing.The normality test,model fitting ability,predictive ability,classification effect were utilized to evaluate the preprocessing methods.The difference of three missing value filled methods was little,through the model test of fitting ability,predictive ability and classification effect.The multiple interpolation method was chosen by intuitive analysis.The results of normality testing showed that the distribution of data was skewed,and the results of normality testing showed that the distribution of data was normally.With the analyzing results of the six normalizing methods,the normalizing method of min–max normalization had the better ability of prediction.The mass metabolomics data obtained by using multiple interpolation and min–max normalization method can process pattern recognition.

关 键 词：代谢组学 数据预处理 缺失值 标准化 

分 类 号：O657.6[理学—分析化学]