Skip to contents

Wrapper function for parameters for the edgeR quasi-likelihood workflow

params_edger_ql.Rd

Parameters for the edgeR quasi-likelihood chain, implemented in Rust via the edge-rs crate and gated against edgeR 4.8.2. Defaults are edgeR's own. The legacy switch picks between two genuinely different pipelines. The current route estimates its own dispersion from the most abundant genes and skips estimateDisp(), which is where most of the runtime went and is edgeR 4's own recommendation. The legacy route shrinks the raw residual deviance, needs a dispersion handed to it, and is the only one where the Poisson bound bites.

Usage

params_edger_ql(
  norm_method = c("TMM", "TMMwsp", "RLE", "upperquartile", "none"),
  filter = TRUE,
  min_mean = 0,
  robust = FALSE,
  legacy = FALSE
)

Arguments

norm_method

String. Library size normalisation. "none" leaves every factor at one, which is what Milo's logMS amounts to. One of c("TMM", "TMMwsp", "RLE", "upperquartile", "none"). Defaults to "TMM".

filter

Boolean. Run filterByExpr() before fitting. Turn this off for anything that is not gene expression, e.g. Milo neighbourhood counts, where the heuristic means nothing. Defaults to TRUE.

min_mean

Numeric. Drop features whose mean count across samples is below this. Applied on top of filter. Defaults to 0.0.

robust

Boolean. Robust empirical Bayes squeezing, giving outlier features their own smaller prior degrees of freedom. Defaults to FALSE.

legacy

Boolean. Take edgeR's pre-4.0 quasi-likelihood pipeline. Defaults to FALSE.

Value

A named list with the following elements:

  • norm_method - String. Library size normalisation. "none" leaves every factor at one, which is what Milo's logMS amounts to. One of c("TMM", "TMMwsp", "RLE", "upperquartile", "none"). Defaults to "TMM".

  • filter - Boolean. Run filterByExpr() before fitting. Turn this off for anything that is not gene expression, e.g. Milo neighbourhood counts, where the heuristic means nothing. Defaults to TRUE.

  • min_mean - Numeric. Drop features whose mean count across samples is below this. Applied on top of filter. Defaults to 0.0.

  • robust - Boolean. Robust empirical Bayes squeezing, giving outlier features their own smaller prior degrees of freedom. Defaults to FALSE.

  • legacy - Boolean. Take edgeR's pre-4.0 quasi-likelihood pipeline. Defaults to FALSE.

References

Chen, Lun and Smyth, F1000Research, 2016