Feature correlation sweep figure — plot_feature_correlation

Produces a 3-panel figure showing how estimator performance changes as within-module feature correlation increases, modelling co-expression modules in a correlated omics panel.

Usage

plot_feature_correlation_sweep(
  sweep_results,
  null_results = NULL,
  file = NULL,
  width = 10,
  height = 12
)

Arguments

sweep_results: Results from sweep_mediation_param("feat_cor", ...): a list with $summary (data frame with columns param_value, method, NDE_bias, NIE_bias, NDE_rmse, NIE_rmse).
null_results: Optional results from sweep_mediation_null_by_conf(..., feat_cor = ...) run with feat_cor swept: a data frame with columns feat_cor, method, NIE_type1. If NULL, Panel C is omitted and the figure has 2 panels.
file: Optional file path to save the figure.
width: Figure width in inches. Default 10.
height: Figure height in inches. Default 12.

Value

A patchwork ggplot object.

Details

Panel A: NDE and NIE bias vs feature correlation strength, faceted by estimand. Panel B: NDE and NIE RMSE vs feature correlation strength, faceted by estimand. Panel C: NIE Type I error vs feature correlation strength.

Examples

# Toy inputs standing in for sweep_mediation_param("feat_cor", ...) output
sweep <- list(summary = expand.grid(param_value = c(0, 0.5),
  method = c("UNADJ", "IV2SLS2", "PGC2Gm")))
sweep$summary$NDE_bias <- rnorm(nrow(sweep$summary), 0, 0.05)
sweep$summary$NIE_bias <- rnorm(nrow(sweep$summary), 0, 0.05)
sweep$summary$NDE_rmse <- runif(nrow(sweep$summary), 0.05, 0.15)
sweep$summary$NIE_rmse <- runif(nrow(sweep$summary), 0.05, 0.15)
null <- expand.grid(feat_cor = c(0, 0.5),
  method = c("UNADJ", "IV2SLS2", "PGC2Gm"))
null$NIE_type1 <- runif(nrow(null), 0, 0.1)
plot_feature_correlation_sweep(sweep, null)