Controlling the number of false discoveries: application to high-dimensional genomic data

摘要: Abstract Researchers conducting gene expression microarray experiments often are interested in identifying genes that differentially expressed between two groups of specimens. A straightforward approach to the identification such “differentially expressed” is perform a univariate analysis group mean differences for each gene, and then identify those most statistically significant. However, with large number typically represented on microarray, using nominal significance levels (unadjusted multiple comparisons) will lead many truly not expressed, “false discoveries.” reasonable strategy situations allow small false discoveries, or proportion identified be discoveries. Although previous work has considered control expected discoveries (commonly known as discovery rate), we show these methods may inadequate. We propose stepwise permutation-based procedures specified confidence actual approximately Limited simulation studies demonstrate substantial gain sensitivity detect even when allowing few one apply new analyze data set consisting measurements 9000 paired tumor specimens, collected both before after chemotherapy 20 breast cancer patients. The described broadly applicable problem which variables any measured differ pre-specified groups.