Question

DESeq2 DESeqDataSetFromTximport design formula

0

Entering edit mode

stephensmith • 0

@stephensmith-13036

Last seen 7.7 years ago

Hi,

I am doing some RNA-seq analysis and am running into the Matrix is not full rank error.

It has 16 samples, 8 different treatments so 2 samples per treatment

If i run this

ddsTxi= DESeqDataSetFromTximport( txi, colData= samples, design= ~condition)

it works but anything other than this, e.g.,

ddsTxi= DESeqDataSetFromTximport( txi, colData= samples, design= ~condition + eya1six1)

give the error.

Would it be best to just go with one column (condition) then paste together whichever factors I want to compare further down the line?

Thanks,

Stephen

deseq2 • 7.5k views

ADD COMMENT • link updated 7.7 years ago by sina.nassiri ▴ 130 • written 7.7 years ago by stephensmith • 0

score 0 · Answer 1 · 2017-05-14

0

Entering edit mode

sina.nassiri ▴ 130

@sinanassiri-10062

Last seen 3.5 years ago

Switzerland/Lausanne

Not sure what you mean by "paste together whichever factors I want to compare further down the line", but the problem with matrix not being full rank is that the columns for condition and eya1six1 are linearly dependent in the design matrix, i.e. condition and eya1six1 are confounding. What's your end goal here?

PS: You can check whether your design matrix is full rank or not using is.fullrank() from the limma package.

ADD COMMENT • link 7.7 years ago sina.nassiri ▴ 130

0

Entering edit mode

I guess what I'd like to do is to be able to compare all samples which are expressing six1, eya1, or six1eya1 with the rest in order to increase my statistical significance.

So when I run contast I can call "six1eya1", "y", "n" or contrast= c("six1", "y", "n") etc.

I'm starting to think it might be easier to just run multiple DESeq2 tests with different design matrices for this.

ADD REPLY • link 7.7 years ago stephensmith • 0

0

Entering edit mode

You can load the data at first with design=~1, and then change the design as you like. After changing the design, you need to re-run DESeq() to have the estimates updated.

ADD REPLY • link 7.7 years ago Michael Love 43k

0

Entering edit mode

Ah very nice. That means I won't have to rerun tximport every time.

Thank you very much Michael.

Finding salmon, tximport, deseq2 a very nice pipeline!

ADD REPLY • link 7.7 years ago stephensmith • 0

0

Entering edit mode

Thanks!

ADD REPLY • link 7.7 years ago Michael Love 43k

0

Entering edit mode

Very new user. Thank you for the great work! I am wondering how can you change the design after you already import date using tximport? Could you give an example? Thank you!

ADD REPLY • link 6.5 years ago lorymoon2012 • 0

0

Entering edit mode

You can use design(dds) <- ~ ...

The important thing is only that you have to re-run DESeq() after you change the design.

ADD REPLY • link 6.5 years ago Michael Love 43k