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
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!
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.