Question

DESEQ2 Grouping vs interaction and dds

0

Entering edit mode

aangajala ▴ 20

@aangajala-12237

Last seen 6.2 years ago

Question #1:

I tried to look the documentation for interaction.But, Unable to understand difference between groups and interaction.So difference between following codes, Is it basically same, different ways of doing it?

dds$group <- factor(paste0(dds$race, dds$sampletype))
design(dds) <- ~ group

Vs

design(dds) <- ~ genotype + condition + genotype:condition

Question #2:

I have this coldata, miRNA expression for ( sampletype(normal and cancer), ER status, her2 status). I want to get one table of results for ER status (positive vs negative ) compared to sampletype and another table by Her2 status compared to sample type.

So my question here is, do I have to have dds twice? of I can produce one dds and then compare later?

basically i have three variables, all examples are based on two.

deseq2 • 1.1k views

ADD COMMENT • link updated 6.2 years ago by Michael Love 41k • written 6.2 years ago by aangajala ▴ 20

0

Entering edit mode

It’s hard to follow your description. Can you post example colData?

ADD REPLY • link 6.2 years ago Michael Love 41k

0

Entering edit mode

> head(coldata)
sampletype race androgen_receptor_statu
TCGA.3C.AAAU.01A.11R.A41G.13 Primary Tumor white positive
TCGA.3C.AALI.01A.11R.A41G.13 Primary Tumor black or african american negative
TCGA.3C.AALJ.01A.31R.A41G.13 Primary Tumor black or african american positive
TCGA.3C.AALK.01A.11R.A41G.13 Primary Tumor black or african american positive
TCGA.4H.AAAK.01A.12R.A41G.13 Primary Tumor white positive
TCGA.5L.AAT0.01A.12R.A41G.13 Primary Tumor white positive
estrogen_receptor_status progesterone_receptor_status
TCGA.3C.AAAU.01A.11R.A41G.13 positive positive
TCGA.3C.AALI.01A.11R.A41G.13 positive positive
TCGA.3C.AALJ.01A.31R.A41G.13 positive positive
TCGA.3C.AALK.01A.11R.A41G.13 positive positive
TCGA.4H.AAAK.01A.12R.A41G.13 positive positive
TCGA.5L.AAT0.01A.12R.A41G.13 positive positive
her2_neu_immunohistochemistry_receptor_status TNBCstatus
TCGA.3C.AAAU.01A.11R.A41G.13 negative
TCGA.3C.AALI.01A.11R.A41G.13 positive
TCGA.3C.AALJ.01A.31R.A41G.13 indeterminate
TCGA.3C.AALK.01A.11R.A41G.13 positive
TCGA.4H.AAAK.01A.12R.A41G.13 equivocal
TCGA.5L.AAT0.01A.12R.A41G.13 negative
>

ADD REPLY • link 6.2 years ago aangajala ▴ 20

0

Entering edit mode

Here is my Coldata, I am trying to verify what I am doing is correct?

ADD REPLY • link 6.2 years ago aangajala ▴ 20

score 0 · Answer 1 · 2018-02-15

0

Entering edit mode

Michael Love 41k

@mikelove

Last seen 35 minutes ago

United States

Do you mean to compare the cancer vs normal difference across ER+ and ER-? If so, then yes an interaction is appropriate and you should use a design ~type + ER + ER:type.

And yes, you would rerun DESeq() with different designs if you then wanted to switch to HER2, etc.

design(dds) <- ...
dds <- DESeq(dds)
res1 <- results(dds)

You can then do:

dds <- removeResults(dds)
design(dds) <- ...

And so on.

ADD COMMENT • link 6.2 years ago Michael Love 41k

0

Entering edit mode

Thank you. What will happen if I do not remove. Then it will just overwrite the dds, so basically it is new dds right? or does it append the dds , I mean mix with the previous one?

ADD REPLY • link 6.2 years ago aangajala ▴ 20

0

Entering edit mode

It makes no difference, it will just print a message saying that it removed the results...

ADD REPLY • link 6.2 years ago Michael Love 41k

0

Entering edit mode

You just made my day saying so :) I did a lot of work yesterday with out removing. So, I was worried.

ADD REPLY • link 6.2 years ago aangajala ▴ 20

0

Entering edit mode

What is the difference between , using group and interaction? Is it going to give same results?

dds <- DESeqDataSetFromMatrix(countData = cts,
colData = coldata,
design = ~ type + ER)
dds$group <- factor(paste0(dds$type, dds$ER))
design(dds) <- ~ group

ADD REPLY • link 6.2 years ago aangajala ▴ 20

0

Entering edit mode

For more details on how different designs differ, you'll need to work with a statistician. The support forum is mostly for developers to help users with software questions, but at some point I have to limit the amount of statistical consulting I do here, or else I wouldn't have any time left.