Question: DESeq2, no replicate
0
gravatar for Mahtetie
23 months ago by
Mahtetie0
Germany
Mahtetie0 wrote:

Hello,

I'm working on a RNAseq dataset with two passages and no replicate!

the STAR-HTSeq output with any of:

dds <- DESeq(ddsHTSeq)
dds <- DESeq(ddsHTSeq, minReplicatesForReplace=Inf, betaPrior=T, modelMatrixType="expanded")
dds <- DESeq(ddsHTSeq, betaPrior=T, modelMatrixType="expanded")
dds <- DESeq(ddsHTSeq, minReplicatesForReplace=Inf)

then something like resDF_DW_P2 <- results(dds, contrast=c("group", "DF1_P2", "DW2_P2"), ref= "EV_P2") gives the DEG list with padj values ~0.99.  

was wondering which one is the correct option to proceed. 

would you think it is correct if I only consider the p-values instead of p-adj values?

Many thanks

Mahtetie

ADD COMMENTlink modified 23 months ago by johnmcma10 • written 23 months ago by Mahtetie0
Answer: DESeq2, no replicate
1
gravatar for Michael Love
23 months ago by
Michael Love25k
United States
Michael Love25k wrote:

It's not surprising that there is no significance with no replicates, this is discussed in this paragraph in ?DESeq:

"Experiments without replicates do not allow for estimation of the dispersion of counts around the expected value for each group, which is critical for differential expression analysis. ... We provide this approach for data exploration only, but for accurately identifying differential expression, biological replicates are required."

You should not use unadjusted p-values.

ADD COMMENTlink written 23 months ago by Michael Love25k
Answer: DESeq2, no replicate
1
gravatar for johnmcma
23 months ago by
johnmcma10
johnmcma10 wrote:

DESeq(2) and edgeR are not intended to be used with designs with less than 3 replicates. For those designs, NOIseq or GFOLD may be better choices, but it's still not a good idea to draw conclusions based on their results.

ADD COMMENTlink written 23 months ago by johnmcma10
1

I agree with Michael Love.

As the senior author of the edgeR project, I can tell you 100% that edgeR was always intended to be used with any design that included any degree of replication. It is certainly not restricted to n=3 or more. I am sceptical that NOIseq or GFOLD would give better performance at low replicate numbers.

You can see edgeR demonstrated in this workflow with n=2 in each group: https://f1000research.com/articles/5-1438

You can see edgeR demonstrated here for the smallest possible design with just three libraries in total (n=1 vs n=2): https://www.degruyter.com/doi/10.1515/sagmb-2017-0010

ADD REPLYlink modified 23 months ago • written 23 months ago by Gordon Smyth38k

For residual degrees of freedom between 1 and 3, there’s nothing in particular about DESeq2 or edgeR which would make it not designed for these sample sizes. We compared favorably to GFOLD in the DESeq2 paper, and to NOISeq in Schurch 2016.

In the DESeq2 paper we describe our implementation of a modification to the dispersion prior variance estimation that is necessary for degrees of freedom in this range.

ADD REPLYlink modified 23 months ago • written 23 months ago by Michael Love25k
Please log in to add an answer.

Help
Access

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.
Powered by Biostar version 16.09
Traffic: 245 users visited in the last hour