Size factors upon collapsing replicates in RNA-seq analysis

Question

[mickey_95]

Hello,

I have a time-course RNA-seq dataset with two biological replicates per condition. To identify genes that are DE throughout time, I used DESeq2 and an LRT test. Now, I would like to identify group of genes that behave similarly throughout time using DEGreport::degPatterns(). As input I use significantly DE genes (as identified by LRT) and the rlog-normalized count matrix. This results in many clusters with some clusters showing very similar patterns and containing only few genes. Increasing the cut-off for genes present in a cluster for it to be reported does not improve this much. Considering that my replicates behave very similar (assessed by PCA and hclust), I was thinking about merging them using DESeq2::collapseReplicates(). Mike Love's reply to a question on the Bioconductor forum reassures me that this might be a good idea (https://support.bioconductor.org/p/114506/). My concern now, however, is regarding the size factors calculated upon collapsing the replicates. From what I understand, the size factor from the first replicate is used. Is this correct? Or are the size factors recalculated for the now merged counts during the rlog-transformation?

I am thankful for any comment or suggestion :)

Answer 1

[ATpoint]

The function you refer to takes the rlog counts directly, so the final transformed counts. You therefore do not have to worry about size factors. If you feel like averaging your replicates makes sense then just average the rlog counts using something like rowMeans per group.

[mickey_95]

True! Argh I was overcomplicating things. Thanks a lot!

[ATpoint]

No worries, happens to all of us once in a while :)