Run the limma linear model chain on a count matrix
rs_limma_voom.Rd![[Experimental]](figures/lifecycle-experimental.svg)
Runs optional filterByExpr -> calcNormFactors -> voomLmFit (or
limma-trend) -> contrasts.fit -> eBayes -> topTable for one
coefficient or contrast, implemented in Rust via the edge-rs crate and
gated against limma 3.66.0.
Arguments
- counts
Integer or double matrix. Raw counts of genes x samples. Must not be normalised or log-transformed.
- design
Numeric matrix. The design matrix of samples x coefficients, including the intercept. Must be full rank.
- lib_size
Numeric vector or NULL. Library size per sample. NULL uses the column sums of
counts. Pass the column sums from before gene filtering to match edgeR, which keeps those on a subsetDGEList.- limma_params
Named list. The limma parameters, see
params_limma_voom(), plus eithercoef(a single 0-indexed(!) design column) orcontrast(column-major weights withn_contrastscolumns).
Value
A list with the following elements, all but features_to_keep
with one entry per kept gene, in input order
features_to_keep - Boolean. Which genes survived the filters. Spans the full gene axis of
counts.log_fc - Log2 fold changes of the tested coefficient or contrast.
ci_lower - Lower end of the 95% confidence interval on
log_fc.ci_upper - Upper end of the 95% confidence interval on
log_fc.ave_expr - Average log2 counts per million.
t_stat - Moderated t statistic.
p_values - Raw p-values.
fdr - Benjamini-Hochberg adjusted p-values.
b_stat - Log-odds of differential expression.