Entering edit mode
ccortes
•
0
@ccortes-13066
Last seen 6.9 years ago
Hi all,
My RNA-seq data come from one genotype with three conditions (A, B and C) made in triplicate. The idea is compare each condition and get the specific differential expressed genes for each condition.
After read the Deseq2 manual and several post I think one possibility is:
dds <- DESeq(dds)
res <- results(dds)
resBvsA <- results(dds, contrast=c("condition","B","A"))
res36vs18 <- results(dds, contrast=c("condition","C","A"))
res36vs24 <- results(dds, contrast=c("condition","C","B"))
# To extract middle point between all groups (genes specifically DE in each time)
results(dds, name="conditionA")
results(dds, name="conditionB")
results(dds, name="conditionC")
Am I correct?
Also I am wondering if is posible compare log2FC(B/A) vs log2FC(C/A). Does someone have suggestions on how to implement this test in DESeq2?
Thanks in advance,
Christian
Hi Michael,
Why does 'betaPrior=TRUE' apply? What is special about these type of coefficients and what exactly is the "comparison of a single condition with a middle point"? I think I have a similar situation in my data.
Thanks,
-R
So, it gets into a difference in how DESeq2 formulates coefficients, and this has changed over time. In the version of DESeq2 that was published in 2014, we used a technique, which I haven't seen elsewhere, which we called "expanded model matrices", this just meant that each level of the factor had its own coefficient in addition to an intercept. And we put a zero-centered prior on the "betas", the coefficients other than the intercept, so they were shrunk toward the intercept point. You can read about this in the DESeq2 paper.
However, the current version of DESeq2 doesn't make coefficients like this. It uses the standard R coefficients which would be: Intecept, B vs A, and C vs A. And you can then contrast any pairs using results(dds, contrast=...). I'm not convinced there is a lot of value in contrasting one level with the "middle", but I wanted to say to the original poster here that it is possible to do the "old style" analysis using the current version of DESeq2 by adding betaPrior=TRUE.