Single cell test data with a planted multicellular programme — generate_dialogue_test

Generates synthetic data DIALOGUE should be able to solve. Every cell type gets its own noise and its own sample-level nuisance factors; only the first feature column and the planted genes carry a shared per-sample latent, so anything found beyond that is spurious.

Usage

generate_dialogue_test_data(
  syn_data_params = params_sc_synthetic_dialogue(),
  seed = 42L
)

Arguments

syn_data_params: List. Contains the parameters for the generation of the synthetic data, see: params_sc_synthetic_dialogue().
seed: Integer. The seed for the generation of the synthetic data.

Value

List with the following items

counts - dgRMatrix with cells x genes of raw counts.
obs - data.table with cell_id, cell_grp, sample_id and cell_quality. The last is pure noise, independent of the planted programme: hand it to dialogue_sc(quality_col = "cell_quality") so the covariate does not carry the signal you are trying to find.
var - data.table with the gene information.
features - Named list of numeric matrices, one per cell type, with the cell identifiers as row names. Hand this straight to dialogue_sc().
latent - Numeric vector. The per-sample latent the planted programme follows, named by sample.
planted - Named list of character vectors. The planted gene identifiers per cell type.

Details

Same generator the Rust integration tests use, so the planted structure cannot drift between the two suites. What differs is the scale: R takes the raw count layer and lets the normal ingestion path log-normalise it, where the Rust tests feed the planted layer straight in.

Cells are laid out contiguously by cell type and, within a cell type, by sample, so every cell type sees every sample. That is the easy case for the method and is what a fixture should be.

References

Jerby-Arnon & Regev, Nature Biotechnology, 2022

Examples

# synthetic data with a multicellular programme planted across samples
data <- generate_dialogue_test_data()
names(data$features)
#> [1] "cell_type_1" "cell_type_2" "cell_type_3"
head(data$planted[[1]])
#> [1] "gene_001" "gene_002" "gene_003" "gene_004" "gene_005" "gene_006"