Default parameters for CellSweep denoising

Mirrors the CellSweep reference implementation's defaults. The pseudocounts (celltype_lambda, ambient_lambda, bulk_lambda) are given on the scale you see here and divided by the gene count internally. Two of these are worth knowing about before you touch anything else. freeze_ambient_profile = TRUE keeps the ambient profile at its empty-droplet estimate, which is the recommended path and the only one where alpha_cap, the repulsion terms and cell-type reassignment are live. And freeze_empties only accepts TRUE: the reference gives empty droplets a cell-type component they have no label for, which indexes past the end of the profile matrix and wraps onto the last cell type.

Usage

params_sc_cellsweep(
  freeze_empties = TRUE,
  freeze_ambient_profile = TRUE,
  init_alpha = 0.9,
  init_beta = 0.1,
  alpha_cap = 0.9,
  repulsion_strength = 1e-04,
  max_frac_gene_repulsion = 0.2,
  celltype_lambda = 50,
  ambient_lambda = 50,
  bulk_lambda = 10,
  eps = 1e-12,
  log_eps = as.numeric("1e-300"),
  max_iter = 2000L,
  del0_ll_tol = 0.001,
  min_ll_tol = 1e-06,
  tol_p = 1e-04,
  tol_f = 1e-04,
  norm_from_rounded = FALSE,
  seed = 42L
)

Arguments

freeze_empties

Boolean. Keep the contamination fraction of empty droplets pinned at 1. Only TRUE is supported, see the description. Defaults to TRUE.

freeze_ambient_profile

Boolean. Keep the ambient profile at its empty-droplet estimate rather than re-estimating it as a mixture over the cell-type profiles. Defaults to TRUE.

init_alpha

Numeric. Starting ambient fraction for every real barcode. With freeze_ambient_profile = TRUE the final result barely depends on it, so it sits at alpha_cap. Defaults to 0.9.

init_beta

Numeric. Starting bulk contamination fraction. Set below init_alpha on purpose: bulk and ambient are not fully separable, so this biases unassignable contamination towards ambient. Defaults to 0.1.

alpha_cap

Numeric. Ceiling on the per-cell ambient fraction before the log-likelihood converges. Barcodes wanting to exceed it are excluded from the cell-type profile update and allowed to switch cell type. Defaults to 0.9.

repulsion_strength

Numeric. Strength of the repulsion pushing cell-type profiles away from the ambient profile. Scales with cluster mass, so it is inert on small data and only bites at realistic cell counts. Defaults to 1e-04.

max_frac_gene_repulsion

Numeric. Ceiling on the fraction of any single profile entry that repulsion may remove. Defaults to 0.2.

celltype_lambda

Numeric. Pseudocount smoothing the cell-type profile update. Higher values give smoother profiles. Defaults to 50.0.

ambient_lambda

Numeric. Pseudocount smoothing the ambient profile estimate. Defaults to 50.0.

bulk_lambda

Numeric. Pseudocount smoothing the bulk profile estimate. Defaults to 10.0.

eps

Numeric. Floor on denominators. Defaults to 1e-12.

log_eps

Numeric. Floor on the argument of log. Defaults to 1e-300.

max_iter

Integer. Hard cap on EM iterations. Defaults to 2000L.

del0_ll_tol

Numeric. Log-likelihood change, as a fraction of the first EM step's change, below which stage one ends and parameter convergence starts being checked. Defaults to 0.001.

min_ll_tol

Numeric. Floor on the adaptive tolerance, relative to the current log-likelihood. Stops del0_ll_tol chasing floating point noise. Defaults to 1e-06.

tol_p

Numeric. Convergence threshold on the maximum row-wise L1 change in the cell-type profiles. Defaults to 1e-04.

tol_f

Numeric. Convergence threshold on the change in the total contamination fraction. Defaults to 1e-04.

norm_from_rounded

Boolean. Derive the normalised layer from the integerised counts rather than the denoised floats. Consistent across the two layers at the cost of the sub-integer signal, which is where CellSweep is most informative. Defaults to FALSE.

seed

Integer. Seed for the stochastic rounding of the denoised counts. Defaults to 42L.

Value

A named list with the following elements:

• freeze_empties - Boolean. Keep the contamination fraction of empty droplets pinned at 1. Only TRUE is supported, see the description. Defaults to TRUE.

• freeze_ambient_profile - Boolean. Keep the ambient profile at its empty-droplet estimate rather than re-estimating it as a mixture over the cell-type profiles. Defaults to TRUE.

• init_alpha - Numeric. Starting ambient fraction for every real barcode. With freeze_ambient_profile = TRUE the final result barely depends on it, so it sits at alpha_cap. Defaults to 0.9.

• init_beta - Numeric. Starting bulk contamination fraction. Set below init_alpha on purpose: bulk and ambient are not fully separable, so this biases unassignable contamination towards ambient. Defaults to 0.1.

• alpha_cap - Numeric. Ceiling on the per-cell ambient fraction before the log-likelihood converges. Barcodes wanting to exceed it are excluded from the cell-type profile update and allowed to switch cell type. Defaults to 0.9.

• repulsion_strength - Numeric. Strength of the repulsion pushing cell-type profiles away from the ambient profile. Scales with cluster mass, so it is inert on small data and only bites at realistic cell counts. Defaults to 1e-04.

• max_frac_gene_repulsion - Numeric. Ceiling on the fraction of any single profile entry that repulsion may remove. Defaults to 0.2.

• celltype_lambda - Numeric. Pseudocount smoothing the cell-type profile update. Higher values give smoother profiles. Defaults to 50.0.

• ambient_lambda - Numeric. Pseudocount smoothing the ambient profile estimate. Defaults to 50.0.

• bulk_lambda - Numeric. Pseudocount smoothing the bulk profile estimate. Defaults to 10.0.

• eps - Numeric. Floor on denominators. Defaults to 1e-12.

• log_eps - Numeric. Floor on the argument of log. Defaults to 1e-300.

• max_iter - Integer. Hard cap on EM iterations. Defaults to 2000L.

• del0_ll_tol - Numeric. Log-likelihood change, as a fraction of the first EM step's change, below which stage one ends and parameter convergence starts being checked. Defaults to 0.001.

• min_ll_tol - Numeric. Floor on the adaptive tolerance, relative to the current log-likelihood. Stops del0_ll_tol chasing floating point noise. Defaults to 1e-06.

• tol_p - Numeric. Convergence threshold on the maximum row-wise L1 change in the cell-type profiles. Defaults to 1e-04.

• tol_f - Numeric. Convergence threshold on the change in the total contamination fraction. Defaults to 1e-04.

• norm_from_rounded - Boolean. Derive the normalised layer from the integerised counts rather than the denoised floats. Consistent across the two layers at the cost of the sub-integer signal, which is where CellSweep is most informative. Defaults to FALSE.

• seed - Integer. Seed for the stochastic rounding of the denoised counts. Defaults to 42L.