Extrapolate transcripts that dictate difference between samples in PCA

Question

[lucatucciarone]

I'm interested in extrapolating what transcripts dictates differences between sample in a PCA from RNA-seq data. This is the PCA of RNA-seq on different samples (A,B,C,D,E,F,G):

Samples are in duplicate and are circled for better understanding. Bold samples are controls, samples connected with arrows are treated samples.

What I want to understand is what are the genes that "moves" A to B (Red Arrow) or A to C (Blue arrow). Given the partially similar trajectory of A to C and D to E the second question would be to check whether some genes in A to C are also responsible of the movement of D to E. This same comparison can be made from comparing A to B and F to G.

Do you happen to know a tool that can allow me to better address this questions?

[lucatucciarone]

I have thought of a strategy to partially address my question:

My observation starts with A to C and D to E going into the same direction.
Now, one way to answer that question would be to: Extrapolate from both PCs (1 and 2) genes that are most variable (and thus dictates the PC1 and 2) Check the expression of those gene in: A Vs C D Vs E and select for genes that has a similar expression pattern.

Do you think that this could be a valid strategy? What tools can you suggest me?

[lucatucciarone]

Someone please!!?!?!?!

Answer 1

[GouthamAtla]

First of all, there are no trajectories in PCA. It's not a graph/network.

You can get the 'n' list of genes that are most variable between any two groups by rowVars function that is part of several R packages. These highly variable genes are the drivers of PCA. Usually top 500 genes are considered for PCA.

[lucatucciarone]

This is very interesting reply, thank you. I'd have one more question on the subject:

Given that there are no trajectory, do you think is legitimate to think that " A to C and D to E" may have some genes that go in the same direction and for that reason their "path/differences" may be similar? In other words, can I at least say that the effect that C has on A may be have some genes overlapping what E does on D? If not, what can I say about the similarity of A to C and D to E? If it is possible to see any

[GouthamAtla]

You may need to do differential analysis on all pairs and try to build a network from differential genes.

[lucatucciarone]

Would you kindly suggest me some papers/tools that explains this pipeline that you are suggesting?

[GouthamAtla]

First you need to accept the existing answer if it clarifies your question.

[lucatucciarone]

Oki! Sorry about that, I am new to the website : )

[lucatucciarone]

any news on this subject?