Simulate an independent reference panel

Draws, for every cell type \(j\), n_individuals \(\times\) n_cells profiles from \(\mathcal{N}_G(\boldsymbol{\mu}_{\cdot j},\boldsymbol{\Sigma}_j)\) and labels each with its individual. With n_cells = 1 every column is the purified profile of one individual, as in the population convolution; with n_cells > 1 every individual contributes n_cells cells, as in a single-cell reference. The generative models of this package have no between-individual layer, so individuals are exchangeable labels.

Benchmark methods that estimate the reference moments (MEAD averages per-individual pseudo-bulks, RNA-Sieve pools the cells) receive this panel; methods that take plug-in moments receive the true \((\boldsymbol{\mu},\{\boldsymbol{\Sigma}_j\})\). Call it with its own seed, distinct from the seed of the bulk samples, so that no bulk is built from a profile of the panel.

Usage

simulate_reference_replicates(
  mean_signature_matrix,
  Sigma,
  n_individuals,
  n_cells = 1L,
  seed = NULL
)

Arguments

mean_signature_matrix

\(G\times J\) mean matrix with gene row names and cell-type column names.

Sigma

\(G\times G\times J\) covariance array.

n_individuals

Number of individuals \(R\ge2\).

n_cells

Profiles per individual and cell type.

seed

Optional seed; the global RNG state is restored afterwards.

Value

A list with reference, a list named by cell type of \(G\times(R\,n_{\mathrm{cells}})\) matrices, and individual, a list named by cell type of integer vectors giving the individual of each column.

See also

simulate_bulk_mixture(), simulate_bulk_from_cells()

Examples

mu <- matrix(
  c(5, 1, 1, 1, 5, 1),
  nrow = 3,
  dimnames = list(paste0("g", 1:3), c("ct1", "ct2"))
)
sigma <- array(
  c(diag(3), diag(3)),
  dim = c(3, 3, 2),
  dimnames = list(rownames(mu), rownames(mu), colnames(mu))
)
panel <- simulate_reference_replicates(mu, sigma, n_individuals = 4, seed = 1)
dim(panel$reference$ct1)
#> [1] 3 4
panel$individual$ct1
#> [1] 1 2 3 4