基于偏最小二乘分析的FDR估计研究

The Research of False Discovery Rate Estimation of Statistical Analysis Based on Partial Least Squares

目的基于偏最小二乘模型(PLS)提出一种新的FDR估计方法,并对其准确性进行验证。方法利用偏最小二乘的vip评分筛选变量,结合permutation方法和后退法对筛选结果进行FDR估计。结果模拟实验表明,在变量之间独立时,PLS-FDR方法和三种单变量估计方法都能准确估计FDR;在变量之间存在线性关系时,PLS-FDR方法估计FDR仍然具有无偏性,而三种单变量分析方法则无法准确地进行估计。实例分析表明,PLS-FDR方法对高维数据分析能够提供重要信息。结论在线性数据结构下,使用本文给出的PLS-FDR方法能够得出多变量FDR估计结果。

Objective To provide a new FDR estimation method based on Partial Least Squares （PLS）and to validate its correction as well. Methods We estimated the FDR of feature selection results based on the vip scores obtained by the Partial Least Squares with the permutation and Step-back technique. Results Simulation experiment proved that the PLS-FDR method and three univariate FDR estimation methods have exact estimation results under the independent structure data. But PLS- FDR method had higher accuracy than three univariate FDR estimation methods in dealing with data having liner relationships. Case study proved that PLS-FDR method can provide important information for high dimensional data analysis. Conclusion PLS-FDR method can estimate the multivariate FDR accurately in the data having liner relationships.