Question: DEseq2 time-series design
1
5 weeks ago by
syednajeebashraf0 wrote:

Hi All,

I have time series study, where colData files look like

      cellLine   dex timeStep batch
200142521     TF-1  CTRL       24     A
200142571     TF-1 IFN-g       24     A
200142621     TF-1  CTRL       48     A
200142661     TF-1 IFN-g       48     A
200142701     TF-1  CTRL       72     A
200142741     TF-1 IFN-g       72     A
700347671     TF-1  CTRL       24     B
700347681     TF-1 IFN-g       24     B
700347691     TF-1  CTRL       48     B
700347701     TF-1 IFN-g       48     B
700347711     TF-1  CTRL       72     B
700347721     TF-1 IFN-g       72     B


The aim of the study is to find Significantly differential Expression gene across Time series 24, 48 and 72 hours. I have two replicates for each Time steps.

I have used below design for analysis. dds <- DESeqDataSetFromMatrix(countData = cts, colData = coldata, design = ~ dex + timeStep + dex:timeStep) and then dds <- DESeq(dds ,test="LRT", reduced = ~ dex + timeStep)

My questions/concerns are: 1. Am I am using correct design Matrix for such time series analysis? 2. Since I am having two replicates as time Period so if the Number of replicates is good for such study? ( I don't have any possibility of adding more replicates.

deseq2 • 104 views
modified 4 weeks ago • written 5 weeks ago by syednajeebashraf0

Thanks Michael,

For the same experiment, I have run for 4 different Cell line. Out of these 4 Cell line, I am able to get gene with padj cutoff for 2 cellline while other 2 cell line, I didn't get any gene to lower then padj cutoff. For this two celline, I got gene with pval criteria but not padj. So what would your suggestion for such cases?

Regards, Najeeb

I have no further suggestions. It is underpowered.

0
5 weeks ago by
Michael Love23k
United States
Michael Love23k wrote:

Yes, that's correct. You could also throw in batch into both full and reduced as an additive effect with no interactions with other coefficients. Having two replicates is not desirable for a number of reasons, including bare practicality that one replicate may be of poor quality. But you can see if you can find large effects with this study nevertheless.