Skip to contents

Calculate the principal component regression on batch

calculate_pcr_sc.Rd

Regresses each embedding dimension on the batch labels and weights the per-dimension R-squared by the variance of that dimension. On its own the number says how much of the embedding variance batch explains. For a corrected embedding, the function also runs it on the uncorrected PCA and reports the scIB comparison (pre - post) / pre: 1 means the batch variance is gone, 0 means nothing changed, negative means it got worse.

Usage

calculate_pcr_sc(object, batch_column, embd_to_use = "pca", .verbose = TRUE)

Arguments

object

SingleCells or SingleCellsSubset class.

batch_column

String. The column with the batch information in the obs data of the class.

embd_to_use

String. Which embedding to compute the PCR on. Defaults to "pca".

.verbose

Boolean. Controls verbosity of the function.

Value

A PcrScores object with the following elements

  • pcr - Variance-weighted R-squared of batch on embd_to_use.

  • pcr_pca - The same on the uncorrected PCA.

  • pcr_comparison - (pcr_pca - pcr) / pcr_pca. NA if embd_to_use = "pca".

  • var_explained - Variance per embedding dimension.

  • r_squared - Batch R-squared per embedding dimension.

  • embedding_used - Which embedding the PCR was computed on.

References

Büttner, et al., Nat. Methods, 2019; Luecken, et al., Nat. Methods, 2022

Examples

# share of PCA variance explained by batch
sc <- demo_single_cells(
  syn_data_params = params_sc_synthetic_data(
    n_cells = 600L, n_genes = 50L, n_batches = 3L
  )
)
calculate_pcr_sc(sc, batch_column = "batch_index")
#> Principal Component Regression (batch)
#>   Embedding: pca | Dimensions: 10
#>   PCR:             0.1414

unlink(sc@dir_data, recursive = TRUE, force = TRUE)