bioconductor-scdesign3

name

bioconductor-scdesign3

description

We present a statistical simulator, scDesign3, to generate realistic single-cell and spatial omics data, including various cell states, experimental designs, and feature modalities, by learning interpretable parameters from real data. Using a unified probabilistic model for single-cell and spatial omics data, scDesign3 infers biologically meaningful parameters; assesses the goodness-of-fit of inferred cell clusters, trajectories, and spatial locations; and generates in silico negative and positi

scDesign3

Workflows

Standard Workflow

Simulate a single-cell RNA-seq dataset with a continuous developmental trajectory from a reference dataset and visualize the simulation quality.

library(scDesign3)
library(SingleCellExperiment)
library(ggplot2)

# Load example data
data("example_count", package = "scDesign3")
data("pseudotime", package = "scDesign3")

# Construct reference SingleCellExperiment
example_sce <- SingleCellExperiment(
  assays = list(counts = example_count, logcounts = log1p(example_count)),
  colData = DataFrame(pseudotime = pseudotime)
)

# Run simulation
set.seed(123)
example_simu <- scdesign3(
  sce = example_sce,
  assay_use = "counts",
  celltype = NULL,
  pseudotime = "pseudotime",
  spatial = NULL,
  other_covariates = NULL,
  mu_formula = "s(pseudotime, k = 10, bs = 'cr')",
  sigma_formula = "1",
  family_use = "nb",
  n_cores = 2,
  usebam = FALSE,
  corr_formula = "1",
  copula = "gaussian",
  DT = TRUE,
  pseudo_obs = FALSE,
  return_model = FALSE,
  nonzerovar = FALSE
)

# Create simulated SingleCellExperiment
simu_sce <- SingleCellExperiment(
  list(counts = example_simu$new_count), 
  colData = example_simu$new_covariate
)
logcounts(simu_sce) <- log1p(counts(simu_sce))

# Visualize and compare
compare_figure <- plot_reduceddim(
  ref_sce = example_sce, 
  sce_list = list(simu_sce), 
  name_vec = c("Reference", "scDesign3"),
  assay_use = "logcounts", 
  if_plot = TRUE, 
  color_by = "pseudotime", 
  n_pc = 20
)
plot(compare_figure$p_umap)

Input: A SingleCellExperiment reference object with pseudotime. Output: A simulated SingleCellExperiment object and a UMAP comparison plot.

When to Use

• To simulate realistic single-cell RNA-seq data with continuous developmental trajectories using scdesign3().
• To generate synthetic count matrices that preserve the correlation structure and marginal distributions of a reference dataset.
• To visualize and compare simulated data against reference data using dimensionality reduction plots via plot_reduceddim().

When NOT to Use

• For simple clustering or differential expression analysis without simulation, use packages like scran or scater directly.

Data Requirements

• A SingleCellExperiment object containing a raw count matrix in its assays (e.g., "counts").
• Cell covariates (such as cell types, pseudotime, or spatial coordinates) stored in the colData of the SingleCellExperiment object.

Key Parameters

• sce: The input SingleCellExperiment reference object.
• assay_use ("counts"): The assay name in sce to use for simulation.
• pseudotime: Column name in colData representing pseudotime.
• mu_formula: Formula for the mean parameter of the marginal distribution (e.g., "s(pseudotime, k = 10, bs = 'cr')").
• sigma_formula ("1"): Formula for the dispersion/variance parameter.
• family_use ("nb"): Distribution family to use (e.g., "nb" for negative binomial).
• copula ("gaussian"): Copula type to model gene correlation (e.g., "gaussian" or "vine").
• n_cores (2): Number of cores to use for parallel computation.

Best Practices

• Set a seed using set.seed() before running scdesign3() and visualization functions to ensure reproducibility of the simulated counts and UMAP layouts.
• To model dispersion varying along pseudotime, set sigma_formula to a spline formula (e.g., "s(pseudotime, k = 5, bs = 'cr')"), keeping in mind this increases computational cost.

Common Pitfalls

• High computational cost when modeling complex dispersion: Setting a complex sigma_formula along pseudotime can significantly increase runtime. Fix: Keep sigma_formula = "1" unless variable dispersion is critical.

Alternatives

• scater: For basic single-cell visualization and QC, but does not perform parametric simulation.
• scran: For normalization and variance modeling, but not for generating synthetic datasets.

Citations

• Song, Dongyuan, and Qingyang Wang. 2026. "scDesign3: A unified probabilistic framework for single-cell and spatial omics simulation."

References

• Homepage: bioconductor.org/packages/scDesign3
• Vignette: https://bioconductor.org/packages/release/bioc/vignettes/scDesign3/inst/doc/scDesign3-quickstart-vignette.html