Permutation-based FDR control via empirical null p-value distribution

Estimates the false discovery rate at each candidate threshold using a matrix of p-values computed under the permutation null, and selects the largest threshold whose estimated FDR is at most alpha. Q-values are assigned as the minimum estimated FDR across thresholds at least as large as each observed p-value.

Usage

permutation_fdr(test_stats, null_stats, alpha = 0.05, labels = NULL)

Arguments

test_stats: Numeric vector of observed p-values (named after the Python sibling argument to keep cross-language parity).
null_stats: Numeric matrix of permutation-null p-values with \(n_{perm}\) rows and \(m\) columns.
alpha: Target FDR level (default 0.05).
labels: Optional character vector of test labels.

Value

A morie_rich_result list with original (raw p-values), adjusted (q-values), rejected, method, alpha, n_rejected, n_tests.

Examples

set.seed(1)
m <- 20; nperm <- 200
p_obs <- c(stats::runif(m - 4), c(1e-4, 1e-3, 1e-3, 5e-3))
p_null <- matrix(stats::runif(nperm * m), nperm, m)
permutation_fdr(p_obs, p_null)
#> Adjusted p-values (permutation_fdr)
#> ===================================
#> Call: method=permutation_fdr, alpha=0.0500, n=20 
#> 
#>   Method          permutation_fdr
#>   alpha           0.05
#>   Tests (n)       20
#>   Rejected        4
#>   Min adjusted p  0
#>   Max adjusted p  0.9415
#> 
#> 4 of 20 hypotheses are rejected at alpha=0.0500. Permutation-null FDR with 200 permutations.