Remove batch effect for differential expression

Question

[Ron]

Hi,

I want to do differential expression between two groups(two batches).For clustering analysis of these samples from two groups,I did batch normalization using COMBAT (sva package)on expression values so as to remove batch effects.

But for differential expression,should I do batch normalization as well before proceeding, as DEseq expects raw counts ? Or is there any way DEseq bioconductor takes into account for different batches and does normalization,before doing differential expression?

Any suggestions,please?

Thanks,

Ron

Answer 1

[WouterDeCoster]

Assuming you did a proper design and treatments are distributed over batches (not all treatment A in batch 1 and treatment B in batch 2 but mixed) you can specify the batch in the design formula and this will be taken into account by DESeq2 and similar tools.

An example of a discussion about a design like that can be found at correcting for a batch in DESeq2 and https://support.bioconductor.org/p/76099/ (among many more)

[Ron]

Hi WouterDeCoster,

I read this post,that I can use Combat batch normalization on Counts and then used the normalized values for differential expression with limma (not EdgeR and DESeq). What do you think on this?

[Kevin Blighe]

Don't use Combat. Go by Wouter's advice, i.e., to include batch in the design model / formula.

If you need even more advice, then take that of one of the core developers of Limma himself:

• Question: remove microarray batch effects using Limma

[kunal368]

Hi WouterDeCoster,

How to deal with batch effects in this situation (all treatment A in batch 1 and treatment B in batch 2)? Specifying batch in the design formula gives error. Is it even possible to remove in that case?

[sontiroy]

Hi @WouterDeCoster

So if our data contains certain groups which are from the same batch matching with the group, we cannot perform batch correction during differential expression?

Like Eg.

sample treatment batch

1 control A

2 control A

3 control A

4 Treat B

5 Treat B

6 Treat B