Skip to contents

Information-Theoretic Metric Learning (ITML) for Single-Cell Data

Source: R/scLearn_model_learning.R
runITML.Rd

Performs stable ITML metric learning for single-cell RNA-seq data with multiple fallback mechanisms to ensure numerical stability. Learns a Mahalanobis distance metric using relative distance constraints with information-theoretic regularization.

Usage

runITML(
  high_varGenes,
  expression_profile,
  sample_information,
  gamma = 0.1,
  max_iter = 50,
  epsilon = 1e-06,
  seed = 1,
  verbose = TRUE
)

Arguments

high_varGenes

Character vector of high-variance gene names

expression_profile

Numeric matrix (genes x cells)

sample_information

Named vector of cell type labels

gamma

Slack parameter controlling constraint strictness (default: 0.1)

max_iter

Maximum number of optimization iterations (default: 50)

epsilon

Convergence threshold for early stopping (default: 1e-6)

seed

Random seed for reproducibility (default: 1)

verbose

Whether to print progress messages (default: TRUE)

Value

A list containing:

  • expression_profile_trans: Transformed data matrix (genes x cells)

  • expression_profile_origin: Original expression matrix

  • trans_matrix: Learned transformation matrix

  • sample_information: Input cell labels

  • convergence: Convergence status information

Examples

if (FALSE) { # \dontrun{
# Example usage:
data("scRNA_example")
itml_result <- runITML(
  high_varGenes = scRNA_example$hv_genes,
  expression_profile = scRNA_example$expr_mat,
  sample_information = scRNA_example$cell_types
)
} # }