A unified approach to false discovery rate estimation

摘要: False discovery rate (FDR) methods play an important role in analyzing high-dimensional data. There are two types of FDR, tail area-based FDR and local as well numerous statistical algorithms for estimating or controlling FDR. These differ terms underlying test statistics procedures employed learning. A unifying algorithm simultaneous estimation both is presented that can be applied to a diverse range statistics, including p-values, correlations, z- t-scores. This approach semipararametric based on modified Grenander density estimator. For other than p-values it allows empirical null modeling, so dependencies among tests taken into account. The inference the model employs truncated maximum-likelihood estimation, with cut-off point chosen according false non-discovery rate. proposed procedure generalizes number more specialized thus offers common framework consistent across In comparative study unified performs par best competing yet alternatives. implemented R "fdrtool" package, available under GNU GPL from http://strimmerlab.org/software/fdrtool/ package archive CRAN.