Skip to contents

Doublet detection with boosted doublet classification

doublet_detection_boost_sc.Rd

This function implements the boosted doublet detection. It generates through several iterations simulated doublets, generate kNN graphs, runs Louvain clustering and assesses how often an observed cells clsuters together with the simulated doublets.

Usage

doublet_detection_boost_sc(
  object,
  boost_params = params_boost(),
  cells_to_use = NULL,
  group_by = NULL,
  seed = 42L,
  streaming = NULL,
  .verbose = TRUE
)

Arguments

object

SingleCells class.

boost_params

A list with the final scrublet parameters, see params_boost() for full details.

cells_to_use

Optional string. Names of the cells to use for the run of the boosted doublet detection. Useful when you wish to run doublet detection on individual batches within your data. The object returned will be specifically using these cells.

group_by

Optional grouping variable. Useful if you want to run the method on a per-sample basis.

seed

Integer. Random seed.

streaming

Optional Boolean. Shall the data be streamed in. Useful for larger data sets where you wish to avoid loading in the whole data. If NULL, will automatically detect.

.verbose

Boolean or integer. Controls verbosity and returns run times. FALSE -> quiet, TRUE or 1L -> normal verbosity, 2L -> detailed verbosity.

Value

A boost_res class that has with the following items:

  • predicted_doublets - Boolean vector indicating which observed cells predicted as doublets (TRUE = doublet, FALSE = singlet).

  • doublet_scores_obs - Numerical vector with the likelihood of being a doublet for the observed cells.

  • voting_avg - Numerical vector with the average voting score.

Examples

# boosted doublet detection over five iterations
sc <- demo_single_cells(prepped = FALSE)
doublet_detection_boost_sc(
  sc,
  boost_params = params_boost(
    hvg = list(min_gene_var_pctl = 0.0),
    pca = list(no_pcs = 10L),
    n_iters = 5L
  ),
  .verbose = FALSE
)
#> BoostRes: 500 cells, 1 doublets (0.2%)
#>   Score range: [0.0125, 0.8289]

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