OPLS方法的原理及其在代谢组学数据判别分析中的应用  

The Theoretical Properties of Orthogonal Projection to Latent Structures( OPLS) and its Application in Metabolomics Data Analysis

作　　者：李俊南[1] 侯艳[1] 孙凤宇[1] 李康[1] 

机构地区：[1]哈尔滨医科大学卫生统计学教研室,150086

出　　处：《中国卫生统计》2014年第5期765-769,共5页Chinese Journal of Health Statistics

基　　金：高等学校博士学科专项基金(项目编号:20122307110004);国家自然科学基金资助(项目编号:81172767)

摘　　要：目的探讨orthogonal projection to latent structures(OPLS)方法的原理、特点及其在代谢组学高维数据分析中的应用。方法通过R语言编程实现OPLS方法,利用模拟试验探索OPLS的特性及适用条件,并通过实际数据进行验证。结果利用一个OPLS预测主成分的模型拟合效果与利用偏最小二乘(PLS)多个主成分的模型拟合效果相同,同时具有较好的判别能力,其得分图的可视化效果优于PLS。结论 OPLS能够有效去除自变量矩阵X中与因变量Y无关的信息,使模型变得简单、易于解释,同时具有较好的可视化效果,可有效地用于代谢组学数据分析中。Objective To explore the theoretical properties and characteristics of orthogonal projection to latent structures（ OPLS） and the application in metabolomics data as well. Methods We independently program OPLS algorithms using R and explore the OPLS method to the real conditions with simulation studies and validated the utility to the real example. Results The model fitting effect with single OPLS principal component is identical to that with multiple components obtained from PLS method. Meanwhile,OPLS has excellent external prediction ability as well as a better visualization effect compared with PLS. Conclusion OPLS can effectively remove information unrelated to the response,which improved,its explanatory ability and visualization effect. This method can effectively be used in metabolomics data.

关 键 词：ORTHOGONAL PROJECTION to LATENT structures(OPLS) 模型拟合 可视化 代谢组学 

分 类 号：R195.1[医药卫生—卫生统计学]