Hi my understanding is that we should not use corrected counts with something like edgeR or limma?

However, because ComBat_seq is correcting at the raw count level and then input into edgeR for TMM follow with voom can I use this for downstream limma or should I just run the voom limma with uncorrected and add the batch as covariate?

thanks in advance!

A

should I just run the voom limma with uncorrected and add the batch as covariate?

This is the preferable option. Also, by including batch as a covariate in a model with formula of, e.g., ~ condition + batch, the test statistics that you derive for condition will be adjusted for the effects of batch.

Following this approach, if you later wish to obtain batch-corrected expression levels for downstream purposes like heatmap + clustering, classification algorithms, et cetera, then use limma::removeBatchEffect() on the output of cpm(y, log = TRUE)

A related question and answers: edgeR and removeBatchEffect()

Kevin