Question

Using varianceStabilizingTransformation (VST) with no replications?

1

Entering edit mode

Sarah_piggy ▴ 10

@nmgduan-19492

Last seen 22 months ago

United States

Dear developers,

Could I use the varianceStabilizingTransformation (VST) to treat RNA-seq data with no replications?

Thanks in advance!

deseq2 • 1.1k views

ADD COMMENT • link updated 5.3 years ago by Michael Love 41k • written 5.3 years ago by Sarah_piggy ▴ 10

score 2 · Answer 1 · 2019-01-16

2

Entering edit mode

Michael Love 41k

@mikelove

Last seen 8 hours ago

United States

You can estimate the parameters on another dataset, and apply them to a new one. See ?varianceStabilizingTransformation

ADD COMMENT • link 5.3 years ago Michael Love 41k

0

Entering edit mode

sorry for bothering you again.

I have only one count dataset of 6 samples without replications. So I couldn't estimate the parameters on another dataset and use dispersionFunction to save the dispersion function. What do you suggest me to do?

I used dataset <- DESeqDataSetFromMatrix(countData = countdata,colData = colData, design = ~ condition, tidy = TRUE, ignoreRank = FALSE) to get a DESeqDataSet.

But when I used VSTdataset <- varianceStabilizingTransformation(dataset, blind = FALSE, fitType = "local"), it showed error in estimateDispersionsGeneEst(object, quiet = TRUE) : the number of samples and the number of model coefficients are equal, i.e., there are no replicates to estimate the dispersion. use an alternate design formula. I thought it might be caused by no replications.

Then I used dataset <- DESeqDataSetFromMatrix(countData = countdata,colData = colData, design = ~ 1, tidy = TRUE, ignoreRank = FALSE) to get another DESeqDataSet. Fortunately, the VSTdataset <- varianceStabilizingTransformation(dataset, blind = FALSE, fitType = "local") worked.

Do you think the design= ~1 and blind = FALSE or TURE can be used for dataset with no replications?

Thanks in advance!

ADD REPLY • link 5.3 years ago Sarah_piggy ▴ 10

0

Entering edit mode

You can run the VST on 6 samples without replication using blind=TRUE (the default) and design=~1.

There's not a good case for end-users to use the ignoreRank argument. This is just for the DEXSeq package, kind of a backdoor.

ADD REPLY • link 5.3 years ago Michael Love 41k

0

Entering edit mode

Thanks for your prompt response!

ADD REPLY • link 5.3 years ago Sarah_piggy ▴ 10

0

Entering edit mode

I also encountered the same problem, thank you for your answer, but I still want to ask a question as a bioinformatic rookie. When I use the above command like VSTdataset <- varianceStabilizingTransformation(dataset, blind = FALSE, fitType = "local") , which parameter for fitType should be selected for dataset without replications? parametric, local or mean?

ADD REPLY • link 5.3 years ago 1610379 • 0

0

Entering edit mode

Any of those are valid

ADD REPLY • link 5.3 years ago Michael Love 41k

0

Entering edit mode

Any of those are valid choices.

ADD REPLY • link 5.3 years ago Michael Love 41k