Plot deconvolution metric heatmaps

For each algorithm, visualises a selected score over the design grid of simulated \(\boldsymbol{p}\) (and related scenario factors). Requires optional Suggests packages ComplexHeatmap, circlize, and viridis (install ComplexHeatmap via Bioconductor). Use this helper only for linked, annotated, hierarchical grids. For algorithm-similarity correlation matrices prefer plot_algorithm_similarity() (ggplot2::geom_tile()).

Usage

plot_correlation_Heatmap(
  distribution_metrics,
  score_variable = "model_mse",
  n_break = 20,
  uni_scale = TRUE,
  file = NULL
)

Arguments

distribution_metrics

Tibble of metric scores from a benchmark.

score_variable

Column name of the metric to display ("model_mse", "model_rmse", "model_coef_determination", "model_coef_determination_adjusted", "model_mae", "model_cor", "model_ccc"). Matching is case-insensitive.

n_break

Number of colour breaks.

uni_scale

If FALSE, each panel uses its own colour scale.

file

Optional PDF path. When supplied, heatmaps are drawn with grDevices::pdf(); a missing .pdf suffix is added (G4.0).

Value

A named list of ComplexHeatmap heatmap objects (one per algorithm).

See also

plot_algorithm_similarity(), plot_mc_metric_dots()

Examples

metrics <- tibble::tibble(
  correlation_celltype1 = c(0, 0, 0.5, 0.5),
  correlation_celltype2 = c(0, 0.5, 0, 0.5),
  algorithm = "nnls",
  model_mse = c(0.01, 0.02, 0.015, 0.03)
)
if (
  requireNamespace("ComplexHeatmap", quietly = TRUE) &&
    requireNamespace("circlize", quietly = TRUE) &&
    requireNamespace("viridis", quietly = TRUE)
) {
  ht <- plot_correlation_Heatmap(metrics, score_variable = "model_mse")
  names(ht)
}
#> [1] "nnls"