Correcting batch effects between RPKM datasets?

Question

[JacobS]

I have two datasets, one with ~250 sample, and another with 7 samples. Both datasets are of RPKM values computed from human RNA-Seq. I don't have access to the primary reads files.

Is there a good way to batch-correct these datasets so that I can combine them and scan for expression signatures? I'm currently using an algorithm that creates a geometric average of the RPKM values for groups of genes that belong in a specific signature in order to compare samples, but the RPKM values of the ~250 sample dataset are on average much higher than the 7 sample dataset.

I've used ComBat in the past for the same predicament but with microarray expression data, and it worked perfectly. I'm looking for something analogous for RPKM expression data.

[Sean Davis]

What are you going to do with the combined data? If you are going to do differential expression analysis, what are the groups?

Answer 1

[Ying W]

Have a look here.

Combat should still work for RNA-seq, it can be found in the SVA package.

You could also have a look at the following packages:

• RUVSeq
• EDASeq

[Sean Davis]

Just note that batch effect correction is not always compatible with the experimental design and questions.

[AB]

Hi,

Thanks. That helps. But I was wondering if the input data in Combat would be just the RPKM data or log2 of the RPKM data