Question: DESeq2 multifactor formula design
0
gravatar for Jane
11 months ago by
Jane0
Jane0 wrote:

Hi,

I am pretty new to DESeq2. I have read the manual and many question posts, but am still not sure if my design is appropriate or not. Could anyone help me with it?

I have samples from 2 tissues (kidney, liver) with 2 treatment condition (control, Drug) at 2 timepoint (1day, 5days). There are 3 replicates for each situation showing as below. 

sample tissue treatment time
1 kidney control 1day
2 kidney control 1day
3 kidney control 1day
4 kidney control 5days
5 kidney control 5days
6 kidney control 5days
7 kidney Drug 1day
8 kidney Drug 1day
9 kidney Drug 1day
10 kidney Drug 5days
11 kidney Drug 5days
12 kidney Drug 5days
13 liver control 1day
14 liver control 1day
15 liver control 1day
16 liver control 5days
17 liver control 5days
18 liver control 5days
19 liver Drug 1day
20 liver Drug 1day
21 liver Drug 1day
22 liver Drug 5days
23 liver Drug 5days
24 liver Drug 5days

 

I'd like to see 1) the effect of the treatment across time in the two tissues and 2)the treatment effect between liver and kidney.

My first design is  

dds <- DESeqDataSetFromHTSeqCount(sampleTable=meta, directory=InFolder, design=~treatment+time+tissue+treatment:time)

dds <- DESeq(dds, test="LRT", reduced=~treatment+time+tissue)

> resultsNames(ddsLRT)
[1] "Intercept"                      "treatment_treatment_vs_control"
[3] "time_5days_vs_1day"             "tissue_liver_vs_kidney"
[5] "treatmentDrug.time5days"

To answer question 1), I use res <- results(dds, contrast = list(c("time_5days_vs_1day", "treatment_Drug_vs_control")), test="Wald"). But I think it is for kidney. How do I get the effect for liver when I don't have an interaction term between tissue and other things? Or should I use design=~treatment+time+tissue+treatment:time+treatment:tissue instead? If it is the case, is it correct for me to use "reduced=~treatment+time+tissue+treatment:tissue"?

I tried both and the results are a little different, which one is correct?

To answer question 2), can I just use the "tissue_liver_vs_kidney" results with "reduced=~treatment+time+tissue"? Or do I need to use "tissue_liver_vs_kidney" with "reduced=~treatment+time+treatment:time"?

Please let me know if my design is appropriate for my purpose.

Thanks a lot!

ADD COMMENTlink modified 11 months ago by Michael Love25k • written 11 months ago by Jane0
Answer: DESeq2 multifactor formula design
0
gravatar for Michael Love
11 months ago by
Michael Love25k
United States
Michael Love25k wrote:

The way I would approach this experiment is to use the following design ~tissue + time + tissue:treatment + tisse:treatment:time.

This gives you coefficents to account for different baselines for the tissues, different baselines for the two time points, different treatment effects in each tissue at the first time point, and then tissue- and treatment-specific changes between the two time points (so four overall, because there are four combinations of tissue and treatment).

You can take a look at the resultsNames(dds) for these, and they will line up with my description above. I'd recommend meeting with a local statistician to discuss these coefficients if you are new to linear models (there is nothing pariticular about DESeq2 here, these are the same coefficients produced by any linear model in R).

ADD COMMENTlink written 11 months ago by Michael Love25k

Hi Michael,

Thanks for your answer! I have another question. I tried a simpler test use different methods (like DESeq2, DESeq) but p-values are different. Is it right? How can I know which ones are correct? Thanks very much!

 

ADD REPLYlink written 11 months ago by Jane0

This is not a well formulated question. You can take my advice above or you can find someone locally to consult with about your analysis. I wouldn’t make arbitrary decisions about your analysis based on the pvalues.

ADD REPLYlink written 11 months ago by Michael Love25k

Okay. Thank you!

ADD REPLYlink written 11 months ago by Jane0
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: 470 users visited in the last hour