Interpreting a PCA plot produced with DESeq2

[feralBiologist]

I've used DESeq2 to analyse a few RNA-Seq samples. I followed pretty closely the manual and using the following code

vsd <- varianceStabilizingTransformation(dds)
data <- plotPCA(vsd, intgroup=c("condition"), returnData=TRUE)
percentVar <- round(100 * attr(data, "percentVar"))
plotPCA <- ggplot(data, aes(PC1, PC2, color=condition)) +
  geom_point(size=3) +
  xlab(paste0("PC1: ",percentVar[1],"% variance")) +
  ylab(paste0("PC2: ",percentVar[2],"% variance")) +
  geom_text(aes(label=names),hjust=0.25, vjust=-0.5, show_guide = F)
ggsave("PCA.pdf", plot = plotPCA)

I created the following PCA plot:

https://picasaweb.google.com/lh/photo/4oyxkL1uVuSM5ugSt2mYKWmkiUW6j8Usj4KOYagHtYg?feat=directlink

What I don't understand is - what are the units that are recorded on the x/y axes? What's their meaning?

[dpryan]

The axes are dimensionless, they have no units.

[feralBiologist]

The axes are dimensionless, they have no units.

Thanks for your reply! So you are saying the units printed on the x/y axes have no meaning? Do you know how can I edit the code to remove them?

[dpryan]

plotPCA <- plotPCA + theme(axis.text.x = element_blank(), axis.text.y=element_blank())

Or something like that. Check the ggplot2 documentation.

[Fatt]

So you are saying the units printed on the x/y axes have no meaning?

They are the first (x-axis) and second (y-axis) principal components, respectively. It's a projection of your data onto a subspace that captures the maximum variance/minimum error. We can likely say they have no units, but they certainly do have meaning.

[feralBiologist]

They are the first (x-axis) and second (y-axis) principal components, respectively. It's a projection of your data onto a subspace that captures the maximum variance/minimum error. We can likely say they have no units, but they certainly do have meaning.

Yes, I know these are the principal components. It is just that DESeq2 prints units on these axes (you can check the link to the plot in my first post) and I could not make any sense of these. I also saw a lot of other PCA plots (presumably produced by other programs) displaying units on the axes so wondered what these are - just do image search on Google for "PCA plot" and you will see a plenty of graphs displaying units on the axes. It is the units that I find confusing, not the axes themselves.

[feralBiologist]

I found the answer to this on CrossValidated: it seems the units denote the raw component scores. As the components are themselves linear combinations of multiple genes it is hard to interpret these raw scores biologically. They are just coordinates in the two-dimensional PC space and are helpful to simply place the individual samples in that space.

[Skiaphrene]

I might be saying this a bit late, but the % of intertia (ie total variance) captured by each PC is a very useful "quality control" measure for PCA and should be included if possible in the figure. The axis units, as said above, are not very interpretable.

BTW, that's a whopping big 1st PC! Nice!

[student-t]

So, how much % of the total variance should be expected?

[dpryan]

@student-t: I replied to this on biostars.