Select Ridge-Stable Transferable Features — select_transferable

Fits ridge-penalised logistic regression models for each cohort and selects features whose coefficients are both large in magnitude and consistent across cohorts. This emphasises biomarkers that deliver transferable signal.

select_transferable_features(
  x_list,
  y_list,
  n_features = 50L,
  lambda = NULL,
  lambda_choice = c("lambda_1se", "lambda_min"),
  min_coefficient = 0,
  require_sign_consistency = TRUE,
  sign_consistency_threshold = 1,
  standardize = TRUE,
  assay = NULL
)

Arguments

x_list: A matrix-like object, SummarizedExperiment, or list of such objects. Rows represent samples and columns represent features.
y_list: A binary response aligned with x_list. Must be a factor with levels c("No", "Yes"). Provide a list when x_list is a list.
n_features: Maximum number of features to return (default 50).
lambda: Optional penalty parameter supplied to glmnet::glmnet(). When NULL, cohort-specific values are chosen via cross-validation.
lambda_choice: When lambda is NULL, choose either "lambda_1se" (default) or "lambda_min" from glmnet::cv.glmnet() as the operating point for coefficient extraction.
min_coefficient: Minimum absolute coefficient required (applied to the cohort-wise minimum magnitude) before a feature is considered transferable.
require_sign_consistency: If TRUE, only retain features whose ridge coefficients meet the sign_consistency_threshold across cohorts. Note that FALSE is equivalent to sign_consistency_threshold = 0: sign agreement is a non-negative fraction, so a zero threshold never filters.
sign_consistency_threshold: Minimum fraction of cohorts that must agree on coefficient sign direction (default 1.0 requires 100% agreement). Set to values like 0.75 to allow partial sign consistency.
standardize: Logical; passed to glmnet::glmnet() to control feature standardisation prior to fitting. Defaults to TRUE.
assay: For SummarizedExperiment inputs, the assay name or index to extract prior to modelling.

Value

A list containing the selected feature identifiers, per-feature scoring metadata, coefficient matrix, and the lambda value used for each cohort.

Details

Features are aligned across cohorts via intersection of shared column names. For more flexible alignment strategies (e.g., majority voting or median imputation), use the feature_alignment parameter in optimize_panel().