How to correct for known and surrogate variables in DESeq2?
1
0
Entering edit mode
Crystal571 • 0
@49c6c837
Last seen 6 months ago
United States

Dear all, We have RNA-seq data of 9 High force versus 9 Low force scallop adductors along with known covariates like Height, Length, Width, Weight (I have cut() them )and Strain. So in DESeq2 I could correct for these covariates as follows:

#We only chose Height, Weight and Strain as covariates because of their higher Pearson correlation with Force.
dds <- DESeqDataSetFromMatrix( countData = expr, colData = subt, design =~Height+Weight+Strain+Group)
dds <- DESeq(dds)
dat  <- counts(dds, normalized = TRUE)
keep  <- rowMeans(dat) > 1
dat  <- dat[keep, ]
mod  <- model.matrix(~Height+Weight+Strain+Group, colData(dds))
mod0 <- model.matrix(~Height+Weight+Strain, colData(dds))
svseq <- svaseq(dat, mod, mod0)
dds$SV1 <- svseq$sv[,1]
dds$SV2 <- svseq$sv[,2]
dds$SV3 <- svseq$sv[,3]

My questions are:

1.The design that corrects for both known and unknown surrogate variables, which one is correct and what leads to the difference of DEGs?

(1) gained 850 DEGs

design(dds) <-~SV1+SV2+SV3+Height+Weight+Strain+Group

or

(2) gained 214 DEGs

design(dds) <-~SV1+SV2+SV3+Group

2.Which one is a better choice for svaseq?

(1)

mod  <- model.matrix(~Height+Weight+Strain+Group, colData(dds))
mod0 <- model.matrix(~Height+Weight+Strain, colData(dds))

or

(2)

mod  <- model.matrix(~Group, colData(dds))
mod0 <- model.matrix(~1, colData(dds))

3.How should I correct the design if I decide the second mod mod0 for svaseq? just like that?

design(dds) <-~SV+Height+Weight+Strain+Group # gained 1187 DEGs

4.If I want to remove batch effect for downstream analysis, did I encode right?

vsd<-vst(dds,blind = F)
covariates = colData(dds)[,c("SV1",'SV2',"SV3")]
assay(vsd) <- limma::removeBatchEffect(assay(vsd), batch1=subt$Height,batch2=subt$Batch,batch3=subt$Strain,covariates = 
covariates,design=model.matrix(~Group, data=subt))

Appreciate for all responses and advice.

Best regards, Crystal571

DESeq2 BatchEffect sva limma • 639 views
ADD COMMENT
0
Entering edit mode

Hey Crystal, did you end up getting advice? I am facing the exact same questions that you are having right now. Would be very grateful if you could share your experience!

ADD REPLY
0
Entering edit mode

To question 3: if you model technical variation using only ~group vs ~1, then it will be correlated with known technical variables, which you would then exclude.

ADD REPLY
0
Entering edit mode
@james-w-macdonald-5106
Last seen 5 hours ago
United States

Your questions are off-topic for this site, which is intended to help people with technical issues rather than statistical questions. You could always ask over on biostars.org, but I would caution against asking for analysis advice from people who know nothing about your data. If you are unsure, it is much better to find someone local who can help.

0
Entering edit mode

OK, thanks a lot for your advice.

ADD REPLY

Login before adding your answer.

Traffic: 514 users visited in the last hour
Help About
FAQ
Access RSS
API
Stats

Use of this site constitutes acceptance of our User Agreement and Privacy Policy.

Powered by the version 2.3.6