We use edgeR for its GLM fitting.

When we present lists of differentially-expressed genes, colleagues ask to see the raw expression data, sometimes for all replicates.   We usually show the counts-per-million of the fitted data: eg.

class(y)[1]          #”DGEList”

fit <- glmFit(y, design, …)

cpm(fit$fitted.values)

These values are the same for all replicates of given sample.

Colleagues then ask to see the standard error of the fitted values.  Can we estimate the standard errors from the object produced by glmFit?  Even an approximation would help.  Grateful for any suggestions.

No.

For more details, check out this post from Gordon:

A: Confidence intervals on edgeR logFC

If you want standard errors, use limma and voom instead.

To plot individual expression values for a given gene, we usually use CPM values from cpm(y) or RPKM values from rpkm(y). Or alternatively cpm(y, log=TRUE, prior.count=3) or rpkm(y, log=TRUE, prior.count=3).

You could attach conventional standard error values to these plots (SE = s/sqrt(n)), although such naive standard errors may misrepresent (usually understate) the true statistical significant between the groups.