Skip to contents

Transfer labels from a Symphony reference to a query via kNN majority vote

transfer_labels_symphony.Rd

kNN majority vote on the Harmony-corrected embeddings. The reference's z_corr is the searchable index; the query's "symphony" embedding is what gets queried. Reference labels come from the reference's stored labels slot — populate it via the label_columns argument to build_symphony_ref() or post-hoc via add_symphony_labels().

Distances on z_corr are typically Euclidean — set knn_params$ann_dist to "euclidean" unless you have a reason to use cosine.

Usage

transfer_labels_symphony(
  reference,
  query,
  label_column,
  knn_params = params_sc_knn(),
  seed = 42L,
  .verbose = TRUE
)

Arguments

reference

SymphonyReference (must have z_corr and stored labels).

query

SingleCells with a "symphony" embedding attached.

label_column

String. Name of a column in the reference's stored labels.

knn_params

List. Output of params_sc_knn().

seed

Integer.

.verbose

Boolean or integer.

Value

A data.table with columns predicted_<label_column> and confidence_<label_column>, in get_cells_to_keep(query) order.

Examples

# kNN label transfer from the reference onto a mapped query
ref <- demo_single_cells(
  syn_data_params = params_sc_synthetic_data(
    n_cells = 500L,
    n_genes = 50L,
    n_batches = 2L
  )
)
symphony_ref <- build_symphony_ref(
  ref,
  batch_column = "batch_index",
  hvg = get_hvg(ref) + 1L,
  harmony_params = params_sc_harmony(k = 10L),
  no_pcs = 10L,
  label_columns = "cell_grp",
  .verbose = FALSE
)
query <- demo_single_cells(prepped = FALSE, seed = 7L)
query <- map_symphony_query(symphony_ref, query = query, .verbose = FALSE)
labels <- transfer_labels_symphony(
  symphony_ref,
  query = query,
  label_column = "cell_grp",
  .verbose = FALSE
)
head(labels)
#>    predicted_cell_grp confidence_cell_grp
#>                <char>               <num>
#> 1:        cell_type_1           0.8666667
#> 2:        cell_type_2           0.9333333
#> 3:        cell_type_3           0.8000000
#> 4:        cell_type_1           0.8000000
#> 5:        cell_type_2           0.8666667
#> 6:        cell_type_3           0.8666667

unlink(c(ref@dir_data, query@dir_data), recursive = TRUE, force = TRUE)