Question: Differences in contrast results while using DESeq2 for one vs all rest combined conditions
1
3 months ago by
shreygandhi199010 wrote:

I have 4 tissue types which I am trying to contrast between different groups using DeSeq2 in order to find genes unique to each tissue type(Let's say tissueA vs all others). I went through some posts: https://support.bioconductor.org/p/86347/ and https://www.biostars.org/p/317843/ and tried the numeric contrast option.

dds_test <- DESeq(dds_gene)
results(dds_test, contrast = c(1, -1/3, -1/3, -1/3))


As, I did not fully understand which number referred to what condition, and tried using the list form in contrast but the commands were throwing an error. I went through the post https://support.bioconductor.org/p/105087/ and tried to follow the same.

dds_test2<- DESeq(dds_gene, betaPrior=T)
resultsNames(dds_test2)

[1] "Intercept"             "tissueA"     "tissueB"  "tissueC"    "tissueD"

results(dds_test2, contrast = list(c("tissueA"),c("tissueB","tissueC","tissueD")), listValues=c(1, -1/3))


This resulted in completely different outputs. Then I tried to check the previous design resultnames using the resultsNames(dds_test) command which showed the following output:

  [1]  "Intercept"          "tissue_B_vs_A"           "tissue_C_vs_A"            "tissue_D_vs_A"


What is the difference between the two approaches? Which way is the correct way of doing this? Have I misinterpreted something?

modified 3 months ago • written 3 months ago by shreygandhi199010
Answer: C: Differences in contrast results while using DESeq2 for one vs all rest combined
1
3 months ago by
Michael Love23k
United States
Michael Love23k wrote:

To compare one vs all the rest, you don’t want an intercept. Use ~0 + condition.

Thanks for the reply. But I still do not understand what the above formulas do. Also, why is it important to remove the intercept in this case?

Take a look at the linear models chapter here:

http://genomicsclass.github.io/book/

In particular the part about contrasts, and also the difference between including an intercept or not in a linear model.