Select cross-technology anchor features on a training pool

A selected anchor must be detected in every required technology class. Selection is based only on `train_expr` and `train_meta`; held-out samples should never be passed here.

Usage

select_cross_tech_anchor_features(
  train_expr,
  train_meta,
  technology_col = "technology",
  required_technologies = NULL,
  min_detection_rate = 0.05,
  n_anchor = 20L
)

Arguments

train_expr

Numeric matrix, samples x features, training pool.

train_meta

Data frame aligned to `train_expr`.

technology_col

Column identifying technology class.

required_technologies

Technology classes required for selection. `NULL` (default) uses the known classes present in training metadata after alias mapping, excluding `unknown`. Fewer than two known classes raises `singlesample_input_error` with code `INSUFFICIENT_TECHNOLOGY_CLASSES`. An explicit vector retains the requested classes, including absent classes (which prevent anchor selection); an explicit single class remains allowed.

min_detection_rate

Minimum per-technology detection rate.

n_anchor

Maximum number of anchors to select.

Value

Data frame with detection rates and `selected_yes`.