Fit a Bayesian region-specific methylation model

Main user-facing entry point. Given a SummarizedExperiment::SummarizedExperiment with a methylation assay and a GenomicRanges::GRanges of predefined regions, BREAD maps probes to regions, summarizes them per sample, and fits Bayesian region-level models to produce posterior probabilities of directional methylation change under the contrast of interest. Regions are classified as hypermethylated, hypomethylated, or inconclusive.

Usage

fit_bread(
  se,
  features,
  design,
  contrast = NULL,
  delta = 0.1,
  prob_cutoff = 0.95,
  min_probes = 3L,
  feature_class_col = NULL,
  summary_fun = c("mean", "median", "weighted_mean", "pc1"),
  assay_name = NULL,
  input_scale = NULL,
  backend = c("conjugate", "brms"),
  prior = NULL,
  ...
)

Arguments

se: A SummarizedExperiment::SummarizedExperiment with a methylation assay.
features: A GenomicRanges::GRanges of user-defined regions.
design: A one-sided formula giving the model design, e.g. ~ group + sex.
contrast: Character coefficient name of interest. If NULL (default), the first non-intercept coefficient is used and a message is emitted.
delta: Effect-size threshold on the M-value scale. Default 0.10.
prob_cutoff: Posterior probability cutoff for classification. Default 0.95.
min_probes: Minimum probes per region. Default 3.
feature_class_col: Column in mcols(features) giving feature class (used by plot_feature_set(); reserved for class-level pooling in M3).
summary_fun: Region summary: "mean" (default), "median", "weighted_mean", or "pc1".
assay_name: Assay name in se. NULL auto-detects.
input_scale: "M" or "Beta". NULL auto-detects from value range.
backend: One of "conjugate" (default) or "brms".
prior: Optional bread_prior() object (conjugate backend only).
...: Additional arguments forwarded to the backend. For backend = "brms", this accepts iter, chains, cores, seed, etc.

Value

A BreadFit object. Use results(), classifications(), or posterior_draws() to inspect.

Minimal call

The typical call is:

fit_bread(se, features, ~ condition)

Everything else has a sensible default. In particular:

contrast defaults to the first non-intercept coefficient,
assay_name auto-detects from the first assay that looks like methylation (prefers "M", "betas", "Beta", "beta"),
input_scale auto-detects from the assay value range ([0,1] → "Beta", otherwise "M").

Backends

Default is an exact conjugate Normal-Inverse-Gamma posterior (fast, no MCMC). backend = "brms" routes to brms::brm() (one compile + per-region updates); MCMC controls iter, chains, cores, seed can be passed through ... to fit_bread_brms().