A direct approach to false discovery rates by decoy permutations
share
The current approaches to false discovery rates (FDRs) in multiple hypothesis testing are usually based on the null distribution of a test statistic. However, all types of null distributions, including the theoretical, permutation-based and empirical ones, have some inherent drawbacks. For example, the theoretical null might fail because of improper assumptions on the sample distribution. In addition, many existing approaches to FDRs need to estimate the ratio of true null hypotheses, which is difficult and unsatisfactorily addressed. In this paper, we propose the target-decoy procedure, a different approach to FDRs for search of random variables different between cases and controls in general case-control study. Our approach is free of estimation of the null distribution and the ratio of true null hypotheses. The target-decoy procedure simply builds on the ordering of hypotheses by some statistic, and directly estimates the false target discoveries using the competing decoy hypotheses constructed by permutation. We prove that this approach can control the FDR. Simulation demonstrates that it is more stable and powerful than two most popular approaches. We also evaluate our approach on a real microarray data for identifying differential gene expression.
READ FULL TEXT
