Fit BREAD summary-mode model via brms (MCMC backend)

Per-region Bayesian linear regression fitted with brms. The Stan program is compiled once on the first region and reused across the remaining regions via update(..., recompile = FALSE), which makes fitting N regions roughly "one Stan compile + N cheap sampling runs".

Usage

fit_bread_brms(
  region_mat,
  coldata,
  design,
  contrast,
  iter = 2000L,
  chains = 4L,
  cores = 4L,
  seed = 1L,
  silent = 2L,
  refresh = 0L
)

Arguments

region_mat

Region-by-sample numeric matrix.

coldata

Sample metadata.

design

One-sided formula.

contrast

Character coefficient name.

iter, chains, cores, seed

MCMC settings passed to brms::brm.

silent

brms::brm's silent level (default 2).

refresh

brms::brm's refresh (default 0, no progress).

Value

A list with the same shape as fit_bread_summary, but each per-region fit carries $draws (a numeric vector of posterior draws for the contrast coefficient) instead of mu_n / Lambda_n_inv / a_n / b_n. posterior_summary and posterior_draws detect the shape automatically.

Priors

v1 uses brms's default priors (weakly informative). The BREAD bread_prior() object is ignored by this backend — it only drives the conjugate backend. Custom brms priors via the brms_prior argument are planned for a future release.