I'm using DESeq2 to evaluate the effect of drug on gene expression from subjects with varying numbers of samples (2-5). Would this be biased?
design attached
I would not use Sample 12,13,23,24,25, and similars... for being technical replicates not biological. But you do not need to have the same number of samples per condition to do nice statistics (since this question is about that, not DeSeq2).
You could collapse your technical replicates [Subject Tissues: 2, 10, 17, 20 etc].
The function is called "collapseReplicates". As long as each Treatment group contains 2/3+ samples (the more, the better), the statistical power should be okay.
As said above, I would remove / collapse / concatenate technical replicates (same subject, same tissue, same treatment), then you end up with 10 samples per group which is more than OK to work with DESeq2. Statistics behind this RNA-seq analysis package are quite strong and efficient even for a low / non equilibrated number of samples. You can even analyze treatment, tissue and their interaction effects at the same time by using the good design. Good luck with your analyses!
Thank you all for your response. These are very helpful suggestions. I have another related question. Is defining drug effects between two separate subject groups (1-10 drug group and 11-20 placebo group) as valid as defining drug effects in a single subject group before and after treatment (paired analysis)?
Do you have the data for both?
You could set that as a separate factor as well in your design matrix. They're both valid, but it depends on the question you're interested in answering.
- Badges
- Science method
- Similar topics
- Biological Science
- Genetics
- Genetic Analysis
- genetic techniques
- Sequencing