Convenience Wrapper for Cross-Platform Transfer — cross_platform

Fits a cross-platform adapter on the source data, adapts source and target matrices, fits or reuses a prediction model, generates predictions on the target platform, and evaluates performance when target labels are available.

Usage

cross_platform_transfer(
  model,
  source_data,
  source_labels,
  target_data,
  target_labels = NULL,
  strategy = "rank",
  reference_features = NULL,
  reference_summary = c("mean", "median"),
  ...
)

Arguments

model: A training function, an `mlr3` learner, or an already fitted model object accepted by [stats::predict()].
source_data: Numeric source matrix with samples in rows and features in columns.
source_labels: Source labels used for model fitting.
target_data: Numeric target matrix with samples in rows and features in columns.
target_labels: Optional target labels for evaluation.
strategy: Adaptation strategy passed to [CrossPlatformAdapter].
reference_features: Optional reference miRNA names for `strategy = "reference"`.
reference_summary: Summary statistic for reference normalization.
...: Additional arguments forwarded to [CrossPlatformAdapter].

Value

A list with the fitted adapter, adapted matrices, fitted model, predictions, optional evaluation metrics, and domain-shift summaries before and after adaptation.

Examples

set.seed(42)
src <- matrix(rnorm(100 * 5), 100, 5)
tgt <- matrix(rnorm(50 * 5, mean = 1.5), 50, 5)
y_src <- rbinom(100, 1, 0.4)
y_tgt <- rbinom(50, 1, 0.4)

fit_glm <- function(x, y) {
  stats::glm(
    y ~ .,
    data = data.frame(y = as.integer(y), x),
    family = stats::binomial()
  )
}

transfer <- cross_platform_transfer(
  model = fit_glm,
  source_data = src,
  source_labels = y_src,
  target_data = tgt,
  target_labels = y_tgt,
  strategy = "rank"
)

transfer$evaluation$auc
#> [1] 0.5909091