Compute Cross-Connectivity: Predicted-Observed ROI x ROI Matrix

Builds an asymmetric ROI x ROI matrix where entry (i, j) is the correlation between the predicted RDM vector of ROI i and the observed RDM vector of ROI j. This captures how well the model-predicted representational geometry in one ROI matches the data-driven geometry in another.

Usage

feature_rsa_cross_connectivity(
  x,
  method = c("spearman", "pearson"),
  adjust = c("none", "double_center", "residualize_mean"),
  return_components = FALSE,
  use = "pairwise.complete.obs"
)

Arguments

x

Either a regional_mvpa_result produced by feature_rsa_model(..., return_rdm_vectors=TRUE) or the tibble returned by feature_rsa_rdm_vectors().

method

Correlation method, one of "spearman" or "pearson".

adjust

Optional adjustment for ROI-level source/target offsets. Use "none" (default) to return the raw ROI x ROI correlation matrix, "double_center" to subtract additive source and target main effects from that matrix, or "residualize_mean" to remove the grand-mean RDM component from predicted and observed ROI vectors before computing the cross-correlation.

return_components

Logical; if TRUE, return a list containing the requested matrix, the raw matrix, the adjusted matrix, and the source and target offset terms estimated from the raw matrix.

use

Missing-value handling passed to cor.

Value

By default, a numeric matrix of dimension n_ROI x n_ROI. Rows correspond to predicted RDM vectors and columns to observed RDM vectors. The matrix is not necessarily symmetric. If return_components = TRUE, a list is returned with elements matrix, raw_matrix, adjusted_matrix, source_offset, target_offset, grand_mean, method, and adjust.

Examples

if (FALSE) { # \dontrun{
res <- run_regional(
  feature_rsa_model(dataset, design, method = "pls", return_rdm_vectors = TRUE),
  region_mask
)
cross_conn <- feature_rsa_cross_connectivity(res, method = "spearman")
cross_dc <- feature_rsa_cross_connectivity(
  res,
  method = "spearman",
  adjust = "double_center"
)
} # }