Optimize Biomarker Panels with Transferability Focus (Deprecated)

Source: R/optimize_panel_transferable.R

Deprecated thin wrapper. Use optimize_panel() directly, with fitness_mode, train_ratio, and val_ratio: the train/validation/ held-out split and the "within_cohort_val" / "within_cohort_rotating" / "loco" fitness strategies this function used to implement are now part of optimize_panel() itself (a held-out share exists whenever train_ratio + val_ratio < 1).

When feature_pool is NULL, this wrapper still partitions the data and runs select_transferable_features() on the training partition only, so the old leakage-safe auto-selection behaviour survives for one release; pass feature_pool explicitly (or call select_transferable_features() yourself and pass its result) to skip that step and call optimize_panel() directly instead.

optimize_panel_transferable(
  x,
  y,
  objectives = define_objectives(metrics = c("sensitivity", "specificity",
    "num_features")),
  max_features = 5L,
  feature_pool = NULL,
  train_ratio = 0.7,
  val_ratio = 0.2,
  np_alpha = 0.15,
  np_delta = 0.05,
  feature_transform = "pairwise_ratios",
  feature_alignment = "intersection",
  constraints = list(),
  algorithm = c("NSGA-III", "NSGA-II"),
  nsga_control = list(),
  assay = NULL,
  seed = 42L,
  regularized = TRUE,
  regularized_alpha = 0.5,
  selection_threshold = "adaptive",
  n_top_features = 50L,
  fitness_mode = c("within_cohort_val", "within_cohort_rotating", "loco"),
  n_val_splits = 10L,
  record_history = FALSE
)

Arguments

x

List of matrix-like objects representing cohorts.

y

List of binary response vectors aligned with x.

objectives

Named list of objective descriptors as returned by define_objectives().

max_features

Maximum number of base biomarkers permitted in a panel.

feature_pool

Optional subset of feature identifiers. If NULL, select_transferable_features() is run on training data only.

train_ratio

Proportion of each cohort for training (default 0.7).

val_ratio

Proportion of each cohort for validation (default 0.2).

np_alpha

Type I error rate for Neyman-Pearson threshold (default 0.15). Stored in the result for use by calibrate_panel().

np_delta

Tolerance for NP threshold selection (default 0.05). Stored in the result for use by calibrate_panel().

feature_transform

Transformation applied to selected base features on-the-fly during optimization (default "pairwise_ratios"). See feature_transform_registry() for options.

feature_alignment

Strategy for aligning features across cohorts (default "intersection").

constraints

Optional list of constraint descriptors.

algorithm

Multi-objective algorithm: "NSGA-III" (default) or "NSGA-II".

nsga_control

Named list of arguments passed to rmoo.

assay

For SummarizedExperiment inputs, assay name or index.

seed

Integer seed for reproducibility (default 42L, as in optimize_panel()). When NULL and feature_pool is also NULL, a seed is drawn and reused for both the feature-selection partition and the optimize_panel() call, so the two partitions match and held-out samples never leak into feature selection.

regularized

Logical; if TRUE (default), use regularized regression.

regularized_alpha

Elastic net mixing parameter (default 0.5).

selection_threshold

Either a fixed numeric threshold in (0,1) for feature selection, or "adaptive" (default) to use gap-based selection that allows variable panel sizes in the Pareto front. See optimize_panel() for details.

n_top_features

Number of top features to select via select_transferable_features() when feature_pool is NULL (default 50).

fitness_mode

Fitness-evaluation strategy: "within_cohort_val" (default), "within_cohort_rotating", or "loco". See optimize_panel() for the description of each.

n_val_splits

Number of rotating train/val splits to pre-compute when fitness_mode = "within_cohort_rotating" (default 10).

record_history

Logical; if TRUE, capture the population, fitness, and NSGA rank at every generation and attach the result to the returned OptimizationResult. Retrieve via nsga_history(). Default FALSE.

Value

An OptimizationResult containing Pareto-optimal solutions and held-out data for downstream calibration. Use summarize_solutions() to inspect solutions, fit_panel() to fit a model, calibrate_panel() for NP threshold calibration, and evaluate_panel() for validation.