What does PCA results signify?

Question

[prithvi.mastermind]

I have plotted two PCA plots showing expression distribution w.r.t all DEGs and immune DEGs. So, what does the difference in PCA plots signify?

A - PCA plot of all DEGs

B- PCA plot of immune DEGs

[GenoMax]

Should be useful:

Answer 1

[Martombo]

To add to what others have commented: these plots are not surprising at all, because (if I understood correctly) you are only using DEGs between tumor and normal samples for the PCA. Is that so? Then of course these genes will show a great separation between your two groups. What is more interesting is to use either all genes, most variable genes or immune genes (if those are of interest for you) to see if there are some overall differences among your samples, what these are and if they separate your groups of interest (tumor, normal).

[swbarnes2]

Yeah, you need to do this with all genes, not just genes that that are different between your groupings of interest. If you had large effects caused by, say, sex, or batch effect, those wouldn't turn up the way you are doing it. Making the PCA with genes you already know are different doesn't really tell you anything at all.

Answer 2

[Istvan Albert]

different simplified explanations exist, each of course carries the risk of oversimplification, here is one such definition

Imagine two matrices NxM, and we want to know whether the first matrix can be deduced from the second one with a shorter formula (say by multiplying it with a vector of N elements). Perhaps some values in the second matrix are a combination of the values in the first plus a random noise. The other values may all be random and unpredictable.

That is the gist of the PCA plot, it shows how strong the "formula's" contribution is relative to the independent contribution.

In an RNA-Seq, the "formula" would represent genes (rows) that all changed in a predictable way under the second condition.

Answer 3

[i.sudbery]

See other's explainations for what a PCA plot mean in general. As for your question - what is the difference between these two PCA plots, and what does that difference signify, I think probably the most obvious answer is "nothing". In both cases your two groups are cleanly seperated on PC1, which accounts for ~70% of the variance. I think it would be difficult make any further inferences that that with these two plots.

[Istvan Albert]

and to add to the above, since the "all DEGs" groups include the "immune DEGs" it is not surprising they look similar

the original poster should compare the immune DEGs to the non-immune DEGs if there is an expectation that the former group separates whereas the latter doesn't - though I would not expect a change there either