Entering edit mode
13 months ago
Zygpy
•
0
I have 99 samples (96 treated, 3 untreated cells) subjected to bulk RNA seq. All the Fastqc parameters look fine + the sequencing depth is pretty similar across all samples. I generated gene counts for all of them with htseq and ran DeSeq2 on it. Now I'm visualizing the results with the dispersion plot & MA plot and it looks totally different from what I expected. Not sure why or what to do? I tried filtering by removing all genes with a total count less than 100, but it didn't help. I also have 30% upregulated genes and only 2% downregulated.
This needs more prefiltering most likely. A filter of sum more than 100 is imo not meaningful at this sample size, it can still retain genes with a single outliers but zero in 99% of samples. Use something like more than 10 counts in 20% of all samples or better use filterByExpr from edgeR before running DESeq(). At this sample size using lfc in lfcShrink to test against a threshold probably makes sense as well. If true DE profile is very assymetric you might need to define controlGenes for normalization or use genes with very large baseMean to still get proper normalization. Comment please if you need clarification. Do PCA as well, see whether there is evidence for batch effects.
What did you expect? Do you have 3 replicates of untreated control, and 96 replicates of the same treatment? Did you expect no gene expression response? Are there any known genes or pathways that respond to the treatment? How do those look across samples? If the treatment is not identical for all 96 samples, how do correlations look? Are the 3 controls tightly correlated and the treatments correlated? Without knowing anything about your system, what organism, nature of the treatment, etc., it's hard to comment. Naively speaking, I would say treatment caused a massive gene expression response, and that could be your answer. If the treatment is too powerful, the answer might not be helpful. Compare an eyeball to a developing embryo and you'll find an MA plot that looks like yours, but it's not a useful experiment because we can already guess that eyeballs and embryos are vastly different. Was your experiment designed to compare 2 varieties of apples? Or apples and oranges? Back to...what did you expect?