Document bayes_t incompatibility with FDR mode

- Update roxygen docs for apply_fdr parameter
- Fix prm-genotype-power vignette to use wilcoxon for FDR example
- Add explanatory note to power-analysis-workflow vignette

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>

Author: danmaclean

R/power.R

@@ -129,7 +129,8 @@ power_analysis.default <- function(distribution, ...) {
 #' @param on_fit_failure How to handle failed fits: "exclude", "empirical", or "lognormal"
 #' @param proportion_threshold Proportion of peptides that must reach target_power (default 0.5)
 #' @param include_missingness If TRUE, incorporate peptide-specific NA rates into simulations
-#' @param apply_fdr If TRUE, use FDR-aware simulation with Benjamini-Hochberg correction
+#' @param apply_fdr If TRUE, use FDR-aware simulation with Benjamini-Hochberg correction.
+#'   Note: not compatible with `test = "bayes_t"` (Bayes factors cannot be converted to p-values)
 #' @param prop_null Proportion of true null peptides (default 0.9 = 90% unchanged)
 #' @param fdr_threshold FDR threshold for calling discoveries (default 0.05)
 #' @param ... Additional arguments (ignored)

man/power_analysis.peppwr_fits.Rd

@@ -643,6 +643,7 @@ print(result_fdr)
 ```
 
 The `apply_fdr` mode simulates entire experiments with a mixture of null and alternative peptides, then applies Benjamini-Hochberg correction before computing power.
+**Note:** FDR-adjusted mode requires frequentist tests (`wilcoxon` or `bootstrap_t`). The `bayes_t` test is not compatible with FDR mode because Bayes factors cannot be converted to p-values for BH correction.
 
 ### Understanding prop_null
 

vignettes/articles/power-analysis-workflow.Rmd

@@ -388,16 +388,18 @@ In targeted proteomics, you typically select peptides expected to change, so `pr
 
 ### Standard vs FDR-Corrected Power
 
+Note: FDR-aware mode requires frequentist tests (Wilcoxon or bootstrap-t) because Benjamini-Hochberg correction operates on p-values. Bayes factors cannot be meaningfully converted to p-values for this correction.
+
 ```{r power-fdr}
-# Standard power (nominal alpha = 0.05)
+# Standard power with Wilcoxon (nominal alpha = 0.05)
 power_nominal <- power_analysis(fits, effect_size = 2, n_per_group = 3,
-                                find = "power", test = "bayes_t",
+                                find = "power", test = "wilcoxon",
                                 apply_fdr = FALSE, n_sim = 100)
 
 # FDR-aware power (BH correction)
 # prop_null = 0.8 means we assume 80% of peptides have no true effect
 power_fdr <- power_analysis(fits, effect_size = 2, n_per_group = 3,
-                            find = "power", test = "bayes_t",
+                            find = "power", test = "wilcoxon",
                             apply_fdr = TRUE, prop_null = 0.8,
                             fdr_threshold = 0.05, n_sim = 100)
 ```
@@ -407,15 +409,15 @@ power_fdr <- power_analysis(fits, effect_size = 2, n_per_group = 3,
 nominal_power <- power_nominal$simulations$peptide_power
 fdr_power <- power_fdr$simulations$peptide_power
 
-cat("Nominal power (no FDR correction):\n")
+cat("Nominal power (Wilcoxon, no FDR correction):\n")
 if (is.numeric(nominal_power) && length(nominal_power) > 0) {
   cat("  Median power:", round(median(nominal_power, na.rm = TRUE), 3), "\n")
   cat("  % peptides > 80% power:", round(mean(nominal_power > 0.8, na.rm = TRUE) * 100, 1), "%\n")
 } else {
   print(power_nominal)
 }
 
-cat("\nFDR-aware power (BH correction, 80% true nulls):\n")
+cat("\nFDR-aware power (Wilcoxon, BH correction, 80% true nulls):\n")
 if (is.numeric(fdr_power) && length(fdr_power) > 0) {
   cat("  Median power:", round(median(fdr_power, na.rm = TRUE), 3), "\n")
   cat("  % peptides > 80% power:", round(mean(fdr_power > 0.8, na.rm = TRUE) * 100, 1), "%\n")