随机森林回归分析及在代谢调控关系研究中的应用

目的探讨随机森林回归处理非线性、具有交互作用数据的性能,并将其应用于高维代谢组学数据的代谢网络变量筛选。方法通过模拟试验验证随机森林回归在具有交互作用和非线性情况下回归分析的效果,同时应用于卵巢良恶性肿瘤鉴别的代谢组学数据分析。结果模拟实验结果显示:对于具有交互作用及其他非线性关系的模拟数据,随机森林回归模型的效果明显优于多元线性回归模型;卵巢癌代谢组学数据分析显示,使用随机森林回归分析能够获得更为理想的结果。结论随机森林回归作为一种非参数回归技术,在一定的样本含量下(如n100),能够在高维数据中有效地分析具有交互作用和非线性关系的数据。

Methodology of Regression by Random Forest and its Application on Metabolomics

ObjectiveEvaluating the property of regression by random forest on analyzing data with non-linearity or interaction,and its application on the feature selection of metabolic network in high dimensional metabonomic data. MethodsValidating the ability of regression by random forest with simulated data with interaction or non-linearity and applying it on discriminating metabolomic data between benign and malignant ovarian cancer. ResultsSimulations showed that the regression by random forest was superior to the multiple linear regression on analyzing data with interaction or non-linear relation.The analysis on real metabolomics data indicated that the regression by random forest achieved better results. ConclusionRegression by random forest,a non-parametric regression technique,would effectively analyze the high dimensional data with interactions or non-linearity when the data with certain sample size(e.g,n>100 ).