Newer version of DESeq2 gives different results
1
0
Entering edit mode
am39 • 0
@am39-10874
Last seen 3.9 years ago

Hi everybody,

I'm using DESeq2 through phyloseq to identify bacterial OTUs that are significantly differently present between treatments. Yesterday I updated R and all my packages. After doing this, I started getting a very different result. Running the same code and with the same data set on two different computers (with and without the package updates) confirmed that the change was indeed due to the updates, though I can't say for certain which update is responsible.

Old versions:

RStudio 0.99.891, R 3.2.3, Phyloseq 1.14.0, DESeq2 1.10.1 : resulted in 65 OTUs significantly different between treatments

New versions:

RStudio 0.99.891, R 3.1.3, Phyloseq 1.16.2, DESeq2 1.12.3 : resulted in only 12 OTUs significantly different between treatments

The code I used was directly from the phyloseq tutorials (side note: really helpful, thank you for all the support!) 

library("DESeq2")
ephdiff <- phyloseq_to_deseq2(data_subs_d, ~treatment)
ephdiffparam <- DESeq(ephdiff)
resparam <- results(ephdiffparam, cooksCutoff = FALSE)

Any idea what the problem might be? 

Thank you! 

Sasha 

 

deseq2 phyloseq • 2.0k views
ADD COMMENT
0
Entering edit mode
@mikelove
Last seen 8 hours ago
United States

You should not expect that results would be identical after updating package versions.

Here's a post where I explain how and why p-values are sensitive to small changes in methods across package versions:

A: Deseq2 results discrepancies because of versions of R

ADD COMMENT
0
Entering edit mode

Thank you! I wasn't expecting identical results, but I was surprised by just how different they were.

If the p-values are so sensitive to changes in methods, how should we take this into account in interpreting results?

ADD REPLY
0
Entering edit mode
In the DESeq2 paper, we suggest to focus on accurate estimates of effect size, which are less simply a function of sample size and more the quantity of interest, the change seen across conditions.
ADD REPLY

Login before adding your answer.

Traffic: 579 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6