Skip to contents

Train PERCEPTIONx models for multiple drugs

Source: R/train.R
train_models.Rd

This function runs the complete PERCEPTIONx training pipeline for a list of drugs. It performs feature ranking, model training with hyperparameter tuning across different k_features values, and selects the best model per drug based on single-cell test performance.

Usage

train_models(
  drug_list = NULL,
  cancer_type = "PanCan",
  exclude_cancer = "PanCan",
  GOI = NULL,
  k_features_values = NULL,
  ncores = 4,
  output_dir = "./models",
  model_type = "glmnet",
  num_folds = 3,
  num_tree = 500,
  seed = 1,
  alpha_gradient = 0.05,
  lambda_gradient = 20,
  lambda_range = c(1e-04, 1),
  cv_method = "cv",
  progress_cb = NULL
)

Arguments

drug_list

Character vector. Names of drugs to train models for. If NULL, uses the 44 FDA-approved drugs from the paper.

cancer_type

Character. Cancer type for training. Default = "PanCan".

exclude_cancer

Character. Cancer type to exclude from training. Default = "PanCan".

GOI

Character vector. Genes of Interest to use as features.

k_features_values

Numeric vector. Feature counts to test during tuning. If NULL, automatically calculated from expression_rnorm dimensions.

ncores

Integer. Number of CPU cores for parallel feature ranking. Default = 4.

output_dir

Character. Directory to save trained model RDS file. Default = "./models".

model_type

Character. Model type: "glmnet" or "rf". Default = "glmnet".

num_folds

Integer. Number of CV folds. Default = 3.

num_tree

Integer. Number of trees for random forest. Default = 500.

seed

Integer. Random seed for training. Default = 1.

alpha_gradient

Numeric. Step size for alpha grid in glmnet tuning. Default = 0.05.

lambda_gradient

Integer. Number of lambda values in glmnet tuning grid. Default = 20.

lambda_range

Numeric vector of length 2. Min and max lambda for tuning grid. Default = c(0.0001, 1).

cv_method

Character. Cross-validation method. Default = "cv".

progress_cb

Function. Optional progress callback called as progress_cb(phase, i, n, drug) where phase is "rank", "train", or "done". Useful for progress bars in interactive sessions.

Value

A named list of trained model objects, one per drug. Also saved as a single RDS file.

Examples

if (FALSE) { # \dontrun{
  load_depmap(read = TRUE)
  available_genes <- intersect(rownames(DepMap$expression_20Q4),
                               rownames(DepMap$scRNA_complete))
  set.seed(123)
  GOI_100 <- sample(available_genes, 100)
  train_models("abemaciclib", "PanCan", "PanCan", GOI_100, ncores = 1)
} # }