Can I take 'organ type' as a batch effect and use ComBat to remove organ dependent variations?

Question

[rodguinea]

I have a RNA-Seq data set (normalized using the DESeq2 library normalization method) consisting 4200 genes and of samples from 22 different species. For each species I have four samples (one for the liver, one for the brain, the kidney and the lung). But I don't want to consider organ dependent variations. What I am looking for are for genes whose expression changes through species, but not through organs. So my question is, can I consider the categorical variable 'Organ type' as a batch effect in order to eliminate its influence with the ComBat function? If it is not advisable to do that, do you have any recommendation?

*Note: Because I only have 122 samples, I don't think it is advisable to split the data set by organ type.

Thanks a lot

Answer 1

[i.sudbery]

If you want to do differential expression to across species while removing the organ effect, you should simply test the design ~ organ + species and look at the species contrast.

However, i think there are other possible issues with what you are planning to do. Count based RNAseq analysis usually assumes that the sequence being measured is the same in all samples, which it would not be here. The most obvious example of this is if genes are different lengths in two different species. If the gene is expressed at the same level in species A and B, but the used transcript is twice as long in species B, then there will be twice as many reads in species B. There might be a way to deal with this via something similar to tximport's effective gene length calculation and offsetting, but there are other such co-founders, such as GC bias, differing mappability etc.

[rodguinea]

Thanks for the insight! I didn't consider some of your observations