Hi everyone, I'm a new research intern who's been tasked with analyzing some RNA-seq data. I'm running DESeq2 analysis through an automated pipeline, however I keep returning massive fold changes on an order which isn't biologically reasonable (ex million, billions, up to the 10^16th power). I've tried gene specific analysis (GSA), which returned reasonable fold changes, but also left me with no FDR values that were statistically valid. Later I also went through my data and ran only the samples with the best expression counts (i.e not a ton of 0's in the data), which also didn't help. Has anyone had this issue before or knows how to help? Thanks!
*Also, if you don't understand my terminology please ask me to explain — I'm really new to this subject so sometimes my phrasing and vocabulary isn't great
Hi Alex P. ,
It is hard to tell what could be wrong in the case you are describing because neither the data nor the code is provided. However, maybe you should start with some sanity checks and exploratory plots of the input data and the output results:
Input data: DESeq2 expects the read counts non-normalized and non-transformed. You can do some filtering by low read count or low quality samples beforehand, but not much more than that.
Experimental design: I would start by running the analysis with simple two group comparisons to check everything is working fine. Then you can move on to more complex designs, if needed. This way you can check the calculated fold change and compare it to the numbers in the count matrix (by eyeballing the numbers or making simple plots).
Model fit: You might want to check the distribution of the p-values and the fold-change even if the numbers are reasonable. Other plots that might help you assess the validity of the analysis can be found in the package vignette (https://bioconductor.org/packages/release/bioc/vignettes/DESeq2/inst/doc/DESeq2.html).
Although the input data of DESeq2 is non-normalised count, it is critical that the normalisation has been properly done when you run DESeq2. But it is not only decided by the DESeq2 script itself, also decided by experiment design, such as if there is enough copy number for each sample, if the samples were run in different batch (batch effect can be one concern). Therefore, you may run the script you are using with one example data set to first check out if the script has been properly written and fulfil all the basic criteria.
- Badges
- Science topic
- Similar topics
- Computer Science and Engineering
- Bioinformatics