Skip to contents

Flag confident cells from a probability matrix

Source: R/model_training.R
compute_custom_labels.Rd

Compute per-cell confidence using a user-supplied decision function or a robust Tukey-style rule: a cell is considered confident if its maximum lineage probability exceeds `max(base_thresh, quantile + iqr_mult * IQR)` for that cell.

Usage

compute_custom_labels(
  prob_mat,
  base_thresh = 0.4,
  quantile_cut = 0.75,
  iqr_mult = 1.5,
  flag_fn = NULL
)

Arguments

prob_mat

Numeric matrix of probabilities (cells x lineages).

base_thresh

Minimum allowable threshold (default 0.4) used only for the default rule.

quantile_cut

Quantile used to set the adaptive threshold (default 0.75) for the default rule.

iqr_mult

Multiplier applied to the IQR (default 1.5) for the default rule.

flag_fn

Optional function taking a numeric vector of probabilities and returning a single logical value. Defaults to the Tukey-style rule above.

Value

Logical vector length `nrow(prob_mat)` indicating confident cells.

Examples

prob_mat <- rbind(
  cell1 = c(Bcell = 0.9, Tcell = 0.1),
  cell2 = c(Bcell = 0.5, Tcell = 0.5)
)
compute_custom_labels(
  prob_mat,
  flag_fn = function(x) max(x) >= 0.8
)
#> [1]  TRUE FALSE