Differential expression on Combat-adjusted FPKM

Question

[mare116]

I have RNA-seq for 62 samples from 4 groups sequenced in 2 known batches, with a clear batch-effect in the PCA. I've derived expression using FPKM, and I've used Combat on log2(FPKM + 1) values. This removes the batch effect successfully, but now I would like to analyze differential gene expression between the 4 groups.

My question/concern: I've read a few people saying that you shouldn't use these Combat-adjusted values for differential expression analysis, but I don't understand why. Can someone please comment on this or explain?

[Ron]

For differential expression,you should use counts. Refer to this post: A: RNA sequencing data batch effect removal

[WouterDeCoster]

An alternative to account for a batch effect problems is to model the batch effect in the design formula of DESeq2/edgeR/limma-voom.