Recursive Hierarchical kNN with BiocParallel Ensemble

Usage

predict_hierarchical_knn_recursive(
  spe,
  hier_ref,
  hc_tree,
  assay_name = "exprs",
  threshold = 0.7,
  agreement_threshold = 0.8,
  k = 5,
  repeats = 5,
  dist_methods = c("pearson", "cosine"),
  BPPARAM = BiocParallel::SerialParam(),
  out_col = "hier_label",
  chunk_size = 1000L,
  unassigned_label = "Unassigned"
)

Arguments

spe: SpatialExperiment object.
hier_ref: List of node-specific references from build_hierarchical_reference.
hc_tree: hclust object defining the hierarchy.
assay_name: Assay to use (default "exprs").
threshold: Confidence threshold (average kNN probability).
agreement_threshold: Minimum % of ensemble members that must agree (0-1).
k: Number of neighbors for kNN.
repeats: Number of bootstrap repeats per distance method.
dist_methods: Vector of methods, e.g., c("pearson", "cosine").
BPPARAM: BiocParallel parameter (default: SerialParam()).
out_col: Column name for final labels.
chunk_size: Integer specifying the number of cells processed per chunk during prediction. Increasing this value may improve speed but requires more memory. Default is 1000L
unassigned_label: character label for Unassigned cells (default "Unassigned")

Examples

# \donttest{
data("cytoGateR_example", package = "cytoGateR")
image_ids <- SummarizedExperiment::colData(cytoGateR_example)$image_name
spe <- cytoGateR_example[, image_ids == image_ids[1L]]
SummarizedExperiment::colData(spe)$cleaned_core_label <-
  SummarizedExperiment::colData(spe)$core_group
SummarizedExperiment::colData(spe)$cleaned_core_label[
  SummarizedExperiment::colData(spe)$cleaned_core_label == "Unassigned"
] <- "Unknown"
hc_tree <- build_lineage_hierarchy(
  spe,
  label_col = "cleaned_core_label"
)
hier_ref <- build_hierarchical_reference(
  spe,
  hc_tree,
  label_col = "cleaned_core_label",
  top_n = NULL,
  unknown_label = "Unknown"
)
spe <- predict_hierarchical_knn_recursive(
  spe,
  hier_ref,
  hc_tree,
  repeats = 1,
  dist_methods = "pearson",
  k = 3
)
#> Starting recursive ensemble classification for 4363 cells (chunk_size=1000)...
head(SummarizedExperiment::colData(spe)$hier_label)
#> [1] "BnTcell"    "Bcell"      "BnTcell"    "Unassigned" "BnTcell"   
#> [6] "CD4"       
# }