Finding outlier replicates of PCA plot

Question

[Microuser]

Hi, I am analysing quality of some of microarray datasets and I've made a PCA plot. The experiment compares the effect of a nutrient concentration at low and high levels at different times (e.g. 72, 0, 3, 8 minutes, etc). I had read that samples of similar perturbation should cluster together (here all lows should be separate from highs?), but I see that they are all mixed. However, the replicates of each group are close. I don't know which group is outlier and if this quality is okay or not? I also included the RMA report and looking at it, I can't decide which replicate should be ignored when I analyse next. They look noisy to me.

Any help is much appreciated!

https://ibb.co/qpgMdcg https://ibb.co/zx6RTjk

Answer 1

[JC]

Your PCA is showing there is a similarity between your replicates (points are closer) also that the High/Low treatment can be separated. This is a PCA, so the PCA1 and PCA2 are only a representation of the 30% of the variance (kinda low btw), you can also check PCA3 and other dimensions. Your second plot is showing that your samples need normalization, try to normalize first and plot again your PCA. At this point I will not remove any replicate from the analysis.