Predict and fill unknown labels using a weighted kNN model

This function takes a SpatialExperiment object and a trained kNN reference to predict labels for "Unknown" or NA cells. It returns the updated SPE and a probability matrix for the predicted cells only.

Usage

predict_unknown_with_knn(
  spe,
  knn_ref,
  assay_name = "exprs",
  label_col = "cutoff_label",
  out_col = "knn_label_filled",
  pred_col = "knn_pred",
  unknown_label = "Unknown",
  unassigned_label = "Unassigned",
  threshold = 0.6,
  k = 5,
  dist_method = "pearson",
  chunk_size = 10000L
)

Arguments

spe

A SpatialExperiment or SingleCellExperiment object.

knn_ref

A list containing the reference data, labels, and features (from train_custom_knn).

assay_name

Name of the assay to use for features (default "exprs").

label_col

The original label column containing "Unknown" cells.

out_col

Output column name for filled labels.

pred_col

Column name for raw kNN predictions.

unknown_label

String identifying missing labels (default "Unknown").

unassigned_label

Label for cells below the confidence threshold.

threshold

Minimum vote proportion (0-1) to accept a label (default 0.6).

k

Number of neighbors.

dist_method

Distance metric ("pearson", "cosine", etc.).

chunk_size

Number of cells to process per chunk for memory management.

Value

A list containing the updated spe and the prob_mat (unknowns only).

Examples

# \donttest{
data("cytoGateR_example", package = "cytoGateR")
image_ids <- SummarizedExperiment::colData(cytoGateR_example)$image_name
spe <- cytoGateR_example[, image_ids == image_ids[1L]]
knn_ref <- train_custom_knn(
  spe,
  label_col = "core_group",
  unknown_label = "Unassigned",
  features = c("CD3e", "CD20", "CD4", "CD8a"),
  cv_folds = 2,
  repeats = 1,
  agreement_thresh = 0.5,
  k = 3,
  seed = 1
)
#> Starting Parallel Weighted kNN (pearson) cleaning on 1378 cells: 1 repeats...
#> Aggregating results from workers...
prediction <- predict_unknown_with_knn(
  knn_ref$spe,
  knn_ref,
  label_col = "cleaned_core_label",
  unknown_label = "Unassigned",
  threshold = 0.5,
  k = 3
)
#> Predicting 3577 unknown cells using pearson kNN...
head(SummarizedExperiment::colData(prediction$spe)$knn_label_filled)
#> [1] "CD8"     "Bcell"   "BnTcell" "CD8"     "BnTcell" "CD4"    
# }