Skip to contents

Run the edgeR quasi-likelihood chain on a count matrix

rs_edger_ql.Rd

[Experimental] Runs optional filterByExpr -> calcNormFactors -> glmQLFit -> glmQLFTest, implemented in Rust via the edge-rs crate and gated against edgeR 4.8.2. legacy = TRUE adds estimateDisp before the fit. The tested axis does not have to be genes: Milo's neighbourhood counts are tested with the same call, with filter = FALSE.

Usage

rs_edger_ql(counts, design, edger_params)

Arguments

counts

Numeric matrix. Raw counts of features x samples. Must not be normalised or log-transformed.

design

Numeric matrix. The design matrix of samples x coefficients, including the intercept. Needs at least two columns, since the null model has to retain one.

edger_params

Named list. The edgeR parameters, see params_edger_ql(), plus either coef (0-indexed(!) design columns to drop from the null model) or contrast (column-major weights with n_contrasts columns).

Value

A list with the following elements, all but features_to_keep with one entry per kept feature

  • features_to_keep - Boolean. Which features survived the filters. Spans the full feature axis of counts.

  • log_fc - Log2 fold changes of the tested coefficient or contrast.

  • log_cpm - Average log2 counts per million.

  • f_stat - The quasi-likelihood F statistic.

  • p_values - Raw p-values.

  • fdr - Benjamini-Hochberg adjusted p-values.

References

Chen, Lun and Smyth, F1000Research, 2016