Skip to contents

Parameters for the per-cell ScType assignment

params_sctype_cells.Rd

Controls the per-cell path of assign_sc_type(): how the raw ScType scores are rescaled, how hard the scores get smoothed over the sNN graph, and where the cut-offs for an Unknown call and for a mixed cluster sit.

Usage

params_sctype_cells(
  alpha = 0.5,
  iterations = 2L,
  tolerance = 1e-04,
  calibration = c("none", "column_z"),
  score_floor = 0.25,
  purity_threshold = 0.9
)

Arguments

alpha

Numeric. Numeric in [0, 1]. Self-retention during smoothing. Each iteration computes alpha * original + (1 - alpha) * neighbour_average. Defaults to 0.5.

iterations

Integer. Integer >= 0. Number of smoothing iterations. 0 disables smoothing. Defaults to 2L.

tolerance

Numeric. Numeric > 0. Convergence tolerance for the smoothing. Defaults to 1e-04.

calibration

String. "column_z" standardises each cell type's score column across cells, which removes the bias towards cell types whose marker sets happen to produce larger scores. One of c("none", "column_z"). Defaults to "none".

score_floor

Numeric. Numeric >= 0. Minimum score for a cell to get a call instead of NA. Defaults to 0.25.

purity_threshold

Numeric. Numeric in [0, 1]. Cluster purity above which the hybrid assignment keeps the cluster-level call. Defaults to 0.9.

Value

A named list with the following elements:

  • alpha - Numeric. Numeric in [0, 1]. Self-retention during smoothing. Each iteration computes alpha * original + (1 - alpha) * neighbour_average. Defaults to 0.5.

  • iterations - Integer. Integer >= 0. Number of smoothing iterations. 0 disables smoothing. Defaults to 2L.

  • tolerance - Numeric. Numeric > 0. Convergence tolerance for the smoothing. Defaults to 1e-04.

  • calibration - String. "column_z" standardises each cell type's score column across cells, which removes the bias towards cell types whose marker sets happen to produce larger scores. One of c("none", "column_z"). Defaults to "none".

  • score_floor - Numeric. Numeric >= 0. Minimum score for a cell to get a call instead of NA. Defaults to 0.25.

  • purity_threshold - Numeric. Numeric in [0, 1]. Cluster purity above which the hybrid assignment keeps the cluster-level call. Defaults to 0.9.

References

Zhou et al., NIPS, 2004.