Should I pre-normalize RNAseq data before combat?

Question

[zhaoliang0302]

Hello,

I want to use combat to remove batch effct. The format of rnaseq data is counts from RSEM. The documention of sva package said data should be normalized before using combat.

So, how to normalize the data? I want to get counts after removing batch effect for subsequent differential analysis (DEseq/edgR). But DEseq and edgR needs raw counts as input, not TPM/FPKM. What should I do?

Best,
Zhao

Answer 1

[Kevin Blighe]

What evidence do you have that a batch effect exists? I would avoid the use of ComBat. You can simply include batch as a covariate in your design formula and perform differential expression analysis in that way.

So, what I am saying is this: you do not have to use ComBat. If you believe that a batch effect exists, include the batch covariate in the design formula.

[zhaoliang0302]

Hello Kevin, Much thanks. I know what you said now. I plot PCA and found a signifiant difference between two batches. After your explanation, I won't use combat . Now the question is I don't know the expression data is count or FPKM, the official documentation is showed in the pictures The maxima of the expression data is 14266614 and there are decimals. I don't know what kind of this data is. Can you help me?