Local false discovery rate via empirical-Bayes two-component mixture

Estimates the local FDR for each test as \(lfdr_i = \pi_0 \, f_0(z_i) / f(z_i)\), where \(z_i = \Phi^{-1}(1 - p_i / 2)\) are two-sided z-scores, \(f_0\) is the standard-normal null density, \(f\) is a kernel density estimate of the observed z-scores, and \(\pi_0\) is the proportion of null hypotheses estimated by the Storey-style cutoff at \(p > 0.5\).

Usage

local_fdr(p_values, pi0_method = "bootstrap", labels = NULL)

Arguments

p_values: Numeric vector of raw p-values in \(`[0, 1]`\).
pi0_method: Pi-zero estimator. Accepted: "bootstrap" (alias for the Storey-style cutoff at 0.5; retained for API parity with the Python sibling).
labels: Optional character vector of test labels.

Value

A data frame with columns p_value, z_score, local_fdr, and (if supplied) label. The data frame additionally carries class morie_rich_result.

Examples

set.seed(1)
p <- c(stats::runif(80),
       stats::pnorm(-abs(stats::rnorm(20, mean = 3))) * 2)
lfdr <- local_fdr(p)
head(lfdr)
#>     p_value   z_score local_fdr
#> 1 0.2655087 1.1134653 0.4653070
#> 2 0.3721239 0.8925020 0.4540127
#> 3 0.5728534 0.5638542 0.4921580
#> 4 0.9082078 0.1152994 0.7779674
#> 5 0.2016819 1.2767743 0.4877259
#> 6 0.8983897 0.1276958 0.7612524