Estimate cross-omic effect covariance by matched-subject bootstrap — bootstrap_effect

Biological subjects are resampled as whole units, preserving their matched measurements across omics. Bootstrap correlations are estimated for each entity and combined with analytic marginal standard errors from `estimate_effects()`: V = diag(SE) R_boot diag(SE). This stabilizes marginal uncertainty while retaining empirically estimated cross-layer dependence.

Usage

bootstrap_effect_covariance(
  data,
  group,
  reference,
  comparison,
  effects = NULL,
  entities = NULL,
  B = 500L,
  seed = 1L,
  min_n = 3L,
  shrinkage = 0.05,
  min_complete = 50L,
  stratified = TRUE
)

Arguments

data: An `omics_braid_data` object.
group: Metadata group column.
reference: Reference group.
comparison: Comparison group.
effects: Optional output of `estimate_effects()`.
entities: Optional entities to bootstrap. By default, entities present in at least two omics.
B: Number of subject-level bootstrap replicates.
seed: Random seed.
min_n: Minimum observations per group within an omic.
shrinkage: Correlation shrinkage toward the identity in [0,1].
min_complete: Minimum usable bootstrap pairs to estimate a correlation.
stratified: Logical; resample subjects separately within reference and comparison groups. This preserves the observed group sizes and is recommended for fixed two-group designs.

Value

Object of class `omics_braid_covariance` containing a covariance matrix per entity.