Faithful port of `generate_simulated_data()`. Counts are sampled with replacement within each cell type according to Dirichlet-generated mixture proportions. Optionally merges a set of cell types into a single unknown class, following Scaden's post-simulation merge semantics.
Usage
scaden_sim_pb(
sc_data,
metadata = NULL,
celltypes = NULL,
celltype_col = NULL,
assay = "RNA",
slot = "counts",
d_prior = NULL,
n = 500,
samplenum = 5000,
seed = NULL,
sparse = TRUE,
sparse_prob = 0.5,
rare = FALSE,
rare_percentage = 0.4,
unknown_celltypes = character(0),
unknown_label = "Unknown",
verbose = TRUE
)Arguments
- sc_data
A `Seurat`, `SingleCellExperiment`, matrix, `data.frame`, or sparse `Matrix` reference.
- metadata
Optional metadata for matrix-like input.
- celltypes
Optional character vector of cell-type labels for matrix-like input. Used only when `metadata` is not supplied.
- celltype_col
Character scalar giving the metadata column holding cell-type labels.
- assay
Character scalar giving the Seurat assay name.
- slot
Character scalar giving the assay slot/layer (Seurat) or assay name (SingleCellExperiment).
- d_prior
Optional numeric vector of Dirichlet concentration parameters. If `NULL`, a symmetric prior of ones is used. Length must equal the number of cell types *before* unknown merging.
- n
Integer scalar. Target cells per pseudobulk before flooring.
- samplenum
Integer scalar. Number of pseudobulks to generate.
- seed
Optional integer scalar for reproducible simulation.
- sparse
Logical scalar. Zero out a subset of cell types in a subset of samples.
- sparse_prob
Numeric scalar in `[0, 1)`. Controls both the proportion of sparse samples and the proportion of cell types zeroed within them.
- rare
Logical scalar. Perturb a subset of cell types to very small fractions in a subset of samples.
- rare_percentage
Numeric scalar in `[0, 1]`. Fraction of cell types treated as rare when `rare = TRUE`.
- unknown_celltypes
Character vector of cell-type labels to merge into a single unknown class. Defaults to `character(0)` (no merging).
- unknown_label
Character scalar used as the merged label.
- verbose
Logical scalar. Print progress.
Value
A named list with `train_x` (samples x genes), `train_y` (samples x cell types), `cell_num`, `celltypes`, `genes`, `elapsed`, and `settings`.
Details
Unknown merging happens **after** simulation: each original cell type gets its own Dirichlet component and is sampled independently, and the proportion columns of the unknown types are then summed into one column. This mirrors Scaden's `y.groupby(y.columns, axis=1).sum()` and differs from relabelling cells before simulation, which would give the merged class a single Dirichlet component and pool its cells uniformly.