Forum:Do you agree with the practice of combing TCGA with GTEx?

Question

[nuorain]

Hi, Because there is no normal control for my TCGA data, I want to use GTEx as normal control to do DEG analysis. I will correct the batch effect using removeBatchEffect function from limma package. I know this practice is controversial. Some friends said these two dataset is not comparable. Do you agree with that? What is your suggestions? Thank you very much for your input.

Answer 1

[i.sudbery]

Your friends are right. You cannot remove batch effects when the batch effect is confounded with the biological condition of interest. removeBatchEffects removes the average differences between two groups of samples. If those two groups are batches, then it will try to remove the batch effect. If those two groups are cancer and normal, it will try to remove the differences between cancer and normal samples. If the batch and the cancer/normal status are the same groups, then try to remove one set of differences (batch), is the same thing as trying to remove the other (cancer/normal). You will either be left with no differences, or if there are differences, you won't know whether they are genuine cancer/normal differences or just a failure to properly remove the batch effect (indeed, the two are the same thing, statistically speaking).

Answer 2

[swbarnes2]

RemoveBatchEffect is not a magic wand. You can't wish for batch effects between labs to magically disappear.