|
Abstract:
|
Many false discovery rate estimation methods often estimated the proportion of genes or features that truly came from the same distribution regardless of disease or control groups, but we have no idea about how accurate the proportion estimation was. To avoid this problem, our new method used only control group data to build up Bootstrap statistic null distribution, and then estimated fdr similarly to the Efron local fdr approach (Efron, 2004). This new method achieved the closest fdr estimations to true fdr values among five different fdr estimation methods in 72 simulation data sets with different settings. These five different local fdr estimation methods were then applied to Golub microarray data sets (Golub et al.,1999). All of these five fdr methods performed well in feature selection for classification and our new method was one of the best two methods.
|
