Tweet
Hi there,
My apologies if this is a stupid question. I've just moved into RNA-seq analysis, and am trying to figure out some issues.
I'm working with a sub-optimal design (the sequencing was done prior to me coming on to this project). I have bone marrow that have been differentiated in two different medias. I have sequence data for three passages of each, and want to compare how close the differentiated cells look between the culture conditions. Effectively I think this means I no biological replicates (but with three technical replicates assuming the effect of passaging is negligible), which I know means I should be weary of any stats I get. I also have some controls samples with biological replicates. All of the samples were sequenced to a depth of ~30 million reads, except for one which had >200 million reads (assuming our sequencing vendor mis-quantified this sample).
When I perform a PCA using DESeq2, all samples cluster pretty closely together, and away from the controls. Our collaborator performed the same analyses using the limma/Voom pipeline and found that the sample with >200 million reads did not cluster with everything else. This only seemed to occur after TMM normalization (without TMM, our data matches).
I'm now not sure if my analysis is correct. Does DESeq2 normalize the data in a fashion similar to TMM? If so is this 'baked in', or should I be calling a particular function. I tried using estimateSizeFactors(dds), which adds a 'normalizationFactors' assay, but it doesn't seem to affect the rlogged data I'm using to generate the PCA.
I also tried to recapitulate the variance plots (Fig 4. from the vignette), but sizeFactors(dds) returns NULL, and I'm not sure what to use to generate values for this attribute.
These data are Salmon-aligned, and imported in to DESeq2 with:
Any suggestions or help would be greatly appreciated!
Thanks!
My apologies if this is a stupid question. I've just moved into RNA-seq analysis, and am trying to figure out some issues.
I'm working with a sub-optimal design (the sequencing was done prior to me coming on to this project). I have bone marrow that have been differentiated in two different medias. I have sequence data for three passages of each, and want to compare how close the differentiated cells look between the culture conditions. Effectively I think this means I no biological replicates (but with three technical replicates assuming the effect of passaging is negligible), which I know means I should be weary of any stats I get. I also have some controls samples with biological replicates. All of the samples were sequenced to a depth of ~30 million reads, except for one which had >200 million reads (assuming our sequencing vendor mis-quantified this sample).
When I perform a PCA using DESeq2, all samples cluster pretty closely together, and away from the controls. Our collaborator performed the same analyses using the limma/Voom pipeline and found that the sample with >200 million reads did not cluster with everything else. This only seemed to occur after TMM normalization (without TMM, our data matches).
I'm now not sure if my analysis is correct. Does DESeq2 normalize the data in a fashion similar to TMM? If so is this 'baked in', or should I be calling a particular function. I tried using estimateSizeFactors(dds), which adds a 'normalizationFactors' assay, but it doesn't seem to affect the rlogged data I'm using to generate the PCA.
I also tried to recapitulate the variance plots (Fig 4. from the vignette), but sizeFactors(dds) returns NULL, and I'm not sure what to use to generate values for this attribute.
These data are Salmon-aligned, and imported in to DESeq2 with:
Code:
txi <- tximport(files, type="salmon", tx2gene=t2g)
dds <- DESeqDataSetFromTximport(txi, sampleTable, ~condition)
dds <- estimateSizeFactors(dds)
rld <- rlog(dds, blind=FALSE)
plotPCA(rld, intgroup=c("condition", "passage" ))
Thanks!