Comparison when Control is in differen batch
1
0
Entering edit mode
rina • 0
@rina-16738
Last seen 14 days ago
Toulouse

I am performing a differential expression analysis between treated and untreated cell lines using DESeq2. The samples were sequenced in two different batches, however, there is a perfect confounding of the conditions and the batches. All controlled samples are in one batch and all treated samples are in the second one. The perfect confounding does not allow to remove batches or even account for them in the formula. I was wondering is it still valid to compare all treated conditions to the control and then compare the regulation of different genes in response to treatment? For example, I would like to check if AKT1 is upregulated in one treatment when compared to control and downregulated in another one when compared to control.

Thank you!

DESeq2 DifferentialExpression • 103 views
0
Entering edit mode

There is not much —if anything— that can be done that is statistically sound. You could make an assumption that there is no difference between batch 1 and batch 2, but this is most likely an erroneous assumption. If you must proceed with the current data, then regard it as preliminary / pilot data and process each batch independently. Then, transform the normalised counts via vst() or rlog(), followed by a final transformation to Z scores. At least on the Z scale, qualitative comparisons across batch are possible, if even to say things such as 'GeneX is expressed in condition 1, but not condition 2'. Absolute Z > 1.96 is equivalent to p=0.05 on a two-tailed distribution. This is still not ideal, though.

0
Entering edit mode
ATpoint ▴ 590
@atpoint-13662
Last seen 7 hours ago
Germany

Not sure which answer you are expecting. The experiment is confounded as you correctly say, hence all DEGs you see could potentially come from the batch. The interpretation is now on you. It comes now down to time considerations, money and feasability whether you will repeat the experiment properly, or just go ahead pretending there was no batch effect. If you proceed consider results with care, try to validate findings.