Skip to contents

Summarize probe-level methylation into region-level values

Source: R/summarize.R
summarize_features.Rd

Collapses probe × sample methylation values into a region × sample matrix, one row per region in the mapping. BREAD models this matrix directly.

Usage

summarize_features(
  se,
  mapping,
  summary_fun = c("mean", "median", "weighted_mean", "pc1"),
  input_scale = c("M", "Beta"),
  assay_name = "M"
)

Arguments

se

A SummarizedExperiment::SummarizedExperiment.

mapping

Data frame from map_probes_to_features().

summary_fun

One of "mean", "median", "weighted_mean", "pc1".

input_scale

"M" or "Beta". "Beta" inputs are converted to M-values before summarization.

assay_name

Assay name in se. Default "M".

Value

A numeric matrix with one row per region (in the order they first appear in mapping) and one column per sample (matching colnames(assay(se, assay_name))). Attributes: summary_fun, input_scale, assay_name.

Summary functions

  • "mean" (default): arithmetic mean of probes per region per sample.

  • "median": per-sample median across probes in the region.

  • "weighted_mean": inverse-variance weighting, where each probe's weight is 1 / max(var_across_samples, 1e-6). Probes with zero variance receive the minimum-variance weight; if all probes have zero variance, weights fall back to uniform (reducing to the plain mean).

  • "pc1": first principal component of the probes (SVD after row-centering). Returned scores are sign-aligned so they correlate positively with the per-sample mean across probes. Scale is abstract — delta loses its M-value interpretation under "pc1".