Reference bootstrap for the single-cell mixture

Resamples the reference that produced \((\tilde{\boldsymbol{\mu}},\boldsymbol{\Sigma}_j)\) and refits deconvolute_single_cell(). resample = NULL uses donors when every cell type has at least two donor labels, and otherwise resamples columns within each cell type. The within-type resample regenerates the bulk from the fitted single-cell law, at the original \((\hat{\boldsymbol{p}},\hat c,\hat\phi)\) and the original moments, and refits it with the resampled moments, so that both the bulk noise and the reference error enter the interval.

Shuffling cell-type names is not a reference bootstrap. It does not redraw cells or donors, and the likelihood is equivariant under a relabeling, so the draw would not be a source of uncertainty. That check remains equivariance_check_decovart() for the population model.

Usage

reference_bootstrap_single_cell(
  y,
  reference_profiles,
  donor_ids = NULL,
  resample = NULL,
  n_boot = 100L,
  level = 0.95,
  regenerate_bulk = NULL,
  ridge = 1e-06,
  epsilon = 1e-06,
  itmax = 40L
)

Arguments

y

Bulk vector of length \(G\).

reference_profiles

Named list of gene-by-cell (or gene-by-replicate) matrices, one per cell type.

donor_ids

Optional list of donor labels, aligned with reference_profiles.

resample

"donors", "cells", or NULL for the automatic rule above.

n_boot

Number of replicates.

level

Confidence level.

regenerate_bulk

NULL regenerates the bulk only for the within-type resample. TRUE or FALSE overrides that rule.

ridge

Ridge passed to the sample covariance, relative to its mean diagonal.

epsilon

Relative function and step tolerances passed to stats::nlminb().

itmax

Maximum number of iterations.

Value

A list with resample, regenerate_bulk, estimates, interval, n_failed, scales, scale_interval and fit; scales and scale_interval also carry the derived rows platform_scale_a and cell_count_n.

See also

reference_bootstrap_decovart(), confint_single_cell()

Other decovart_inference: bootstrap_decovart(), bootstrap_single_cell(), boundary_diagnostics(), chi_bar_square_pvalue(), confint_decovart(), confint_ilr_backtransform(), confint_profile_decovart(), confint_profile_single_cell(), confint_single_cell(), equivariance_check_decovart(), fit_firth_decovart(), fit_firth_single_cell(), lrt_decovart(), lrt_single_cell(), multistart_decovart(), reference_bootstrap_decovart(), restricted_mle_decovart(), restricted_mle_single_cell()

Examples

set.seed(1)
mu <- cbind(ct1 = c(2, 2, 0.2, 0.2), ct2 = c(0.2, 0.2, 3, 3))
sigma <- array(rep(diag(0.05, 4), 2), c(4, 4, 2))
refs <- lapply(seq_len(2), function(j) {
  t(MASS::mvrnorm(n = 6, mu = mu[, j], Sigma = sigma[, , j]))
})
names(refs) <- colnames(mu)
y <- 20 * drop(mu %*% c(0.7, 0.3)) + rnorm(4, sd = 0.2)
reference_bootstrap_single_cell(y, refs, n_boot = 4)$resample
#> [1] "cells"