CatBoost PRS utilities — catboost_prs • leo.gwas

Train, apply, and rank CatBoost polygenic risk score (PRS) models.

Usage

catboost_prs(a1_matrix, divide = FALSE, divide_ratio = 0.5)

catboost_prs_target(a1_matrix, model)

catboost_prs_rank(
  model,
  pool,
  pool_df,
  types = c("FeatureImportance", "ShapValues", "Interaction"),
  top_k = NULL
)

Arguments

a1_matrix: A1 matrix (with PHENOTYPE and ID columns).
divide: Logical; split into train/val sets. Default FALSE.
divide_ratio: Numeric; train proportion if divide=TRUE. Default 0.5.
model: Trained CatBoost model.
pool: CatBoost pool.
pool_df: Data used to build pool.
types: Character; any of "FeatureImportance","ShapValues","Interaction".
top_k: Integer; top features to keep. Default NULL.

Value

Each function returns a list; contents depend on task (training, target prediction, ranking).

Details

These functions require installed catboost. Training with divide = TRUE also requires caret for stratified data splitting.

See also

catboost::catboost.train(), catboost::catboost.get_feature_importance()