deseq2 batch effect correction

Question

[Devbio]

I am very new to R and doing differential expression analysis using DeSeq2 package. My raw counts of cancer and controls samples are from two different papers, do i need to correct batch effects ? If yes, then which tool will be the best, combat_seq or svaseq ? Please help

[rfran010]

I understand this as you have cancer sample from one paper and control samples from a second paper? Batch correction may not be able to overcome such a difference.

If you mean you have some cancer and controls from one paper and cancer and controls from a second paper, a paired design may make more sense than batch correction.

I would say more information and data exploration, and some deep thought, is needed on if this data and comparison can really tell you anything.

[Devbio]

Yes, I have cancer samples from one paper and control samples from a second paper, is there no way to remove batch effect ? even svaseq wont help ?

[rfran010]

The main question to answer would be how do you know batch correction worked properly and did not introduce artefacts and overt biases? I've not generally gotten good results with batch correction. In my understanding though, batch correction is useful when integrating larger datasets or integrating many different datasets, and is best when you have a reference sample that should be "equal" between different data sets.

If only comparing two conditions, I think it's better to perform more manual and custom analysis.

In general, you will want to start with only genes that are detected in both conditions and then see how it looks on a correlation plot.