Compute Normalized Correlation from Ground Truth Scores — compute_ground_truth

Computes the normalized correlation using user-provided cell scores (e.g., ground truth scores from simulation) and the pre-calculated kernel matrix from a CoPro object.

Usage

compute_ground_truth_ncorr(
  object,
  scores_ct1,
  scores_ct2,
  cellType1,
  cellType2,
  sigma = NULL,
  tol = 1e-04
)

Arguments

object: A CoPro object with kernel matrices computed
scores_ct1: Numeric vector of scores for cell type 1. Must be in the same order as cells in the CoPro object.
scores_ct2: Numeric vector of scores for cell type 2. Must be in the same order as cells in the CoPro object.
cellType1: Name of cell type 1
cellType2: Name of cell type 2
sigma: Sigma value for kernel matrix (default: uses sigmaValueChoice)
tol: Tolerance for SVD calculation (default: 1e-4)

Value

Normalized correlation value

Details

This function is useful for comparing:

Ground truth normalized correlation (from simulated scores)
CoPro's optimized normalized correlation (from CCA)
Permutation distribution

Under null simulation (no cross-type coordination), the ground truth normalized correlation should be close to 0, while CoPro's optimized value will be higher because it searches for the maximum correlation.

Examples

if (FALSE) { # \dontrun{
# After running CoPro analysis on simulated data
# Get ground truth scores from metadata
meta <- br@metaDataSub
gt_scores_A <- meta$smoothed_score[meta$cell_type == "A"]
gt_scores_B <- meta$smoothed_score[meta$cell_type == "B"]

# Compute ground truth normalized correlation
gt_ncorr <- compute_ground_truth_ncorr(
  object = br,
  scores_ct1 = gt_scores_A,
  scores_ct2 = gt_scores_B,
  cellType1 = "A",
  cellType2 = "B"
)
} # }