Pivot Monte Carlo proportion estimates to a long table

Turns the optimisation block of run_simulation_benchmark() into one row per replicate, algorithm, and cell type, with the matching true proportion from theta_true. The raincloud intervals drawn from this table are empirical Monte Carlo quantiles of \(\hat{\boldsymbol{p}}\), not confidence intervals for \(\boldsymbol{p}\) (Allen et al. 2019) .

Usage

pivot_mc_estimates(benchmark)

Arguments

benchmark

List returned by run_simulation_benchmark().

Value

A tibble with algorithm, cell_type, estimate, p_true, error, sample_id, and any scenario metadata columns.

References

Allen M, Poggiali D, Whitaker K, Marshall TR, van Langen J, Kievit RA (2019). “Raincloud Plots: A Multi-Platform Tool for Robust Data Visualization.” Wellcome Open Research, 4, 63. ISSN 2398-502X. doi:10.12688/wellcomeopenres.15191.2 . https://wellcomeopenresearch.org/articles/4-63.

See also

plot_mc_raincloud(), plot_mc_forest()

Examples

set.seed(1)
genes <- paste0("g", 1:2)
cts <- paste0("ct", 1:2)
mu <- matrix(c(20, 22, 22, 20), nrow = 2, dimnames = list(genes, cts))
Sigma <- array(
  c(1, 0, 0, 1, 1, 0, 0, 1),
  dim = c(2, 2, 2),
  dimnames = list(genes, genes, cts)
)
out <- run_simulation_benchmark(
  tibble::tibble(
    ID = "B1",
    true_theta = list(list(p = c(0.5, 0.5), mu = mu, sigma = Sigma))
  ),
  deconvolution_functions = list(
    "nnls" = list(FUN = deconvolute_ratios_nnls)
  ),
  n = 2,
  cores = 1
)
nrow(pivot_mc_estimates(out))
#> [1] 4