Recently I got an RNA-Seq project where the data looks like as in the PCA plot below. In essence the variation between biological replicates is larger than the variation in the tested conditions. Likely the wet-lab has processes each replicate of all conditions at different days. Does it make sense to continue with dif. expression analysis?
•
written
by
grant.hovhannisyan
If I understand the design correctly there are 4 biological replicates of 4 conditions. Additionally, there are 4 batches with each batch consisting of 1 sample from each condition.
If the above holds true, you may be able to mitigate the batch effect by including batch as a covariate in the differential expression model (just make sure they are factors and not numeric).
To get an approximate idea of whether this has a chance of clearing up your batch effect your can use the removeBatchEffect function from limma prior to calculating and plotting the PCA.
you are right regarding the experimental design. I had in mind including batch in the design, but on the other hand it kind of will cancel out the variance between biological replicates, and I am not sure whether the resulting analysis will be meaningfull. Any thoughts?
Another answer from the DESeq2 developer confirming what rpolicastro already said (~replicate+condition), see: https://support.bioconductor.org/p/9152334/#9152369
Somone already sunk money into the experiment, so yeah, you ought to at least generate DEGs. But I think the conclusion from this is that your condition is weaker than the batch effects. You aren't going to get much, and I wouldn't try to put heroic amounts of effort trying. If the submitters really think there should be a signal, they need to redo the experiment as to curb the batch effects, and have way more replicates to get more power.
Log in to answer this question.
