Hi all, I have tried using the DESeq2 plotPCA function after vst normalization of my dds object. However, the PCA plot I have obtained were samples separated on a single principal component PC1 with 100% variance explained. How is this possible and is it even appropriate to proceed with DEG analysis from here?
It means one of two things: 1) One of your features (or some linear transformation of it) is sufficient to explain all your data. 2) You have a feature in your data with greater numeric spread than all others, so all other features are squeezed into a small range and their contributions end up being insignificant.
Option #1 is possible, but option #2 is more likely. If you haven't done so already, I suggest you try normalizing the data so all variables end up on a comparable scale.
Thank you, does this mean that I can still proceed with the DEG analysis, but perhaps further normalization is required for PCA? I have used the normalization that comes with DESeq2 via the vst function and only specified the treatment condition in my design formula. Could that be the issue?
Those are very good observations. However, thinking in biological terms, option 2 will mean that the expression of one gene is varying in an extremely high range compared to all others - a somewhat unrealistic expectation as genes are not expressed in isolation. Though, it could be also possible that the data is not in log-scale, which is createing unwanted variance.
PS: I guess, you meant standardization of variables, instead of normalization.
Log in to answer this question.
This is intriguing me. Is it possible that you could share your vst-normalized data that you are using for the PCA?
sure, here is the drive link to it
Thanks for the data. This doesn't look like a realistic data to me as all the genes seem (sorry I'm on mobile, so only had a quick look) to be expressed at almost the same level (9-11). Nevertheless, unless there is a distant 'outlier' in the data, I don't see how PCA can give 100% variance in first component.
There is quite a significant portion of significantly upregulated/downregulated genes after DESeq2 though... I have approximately 700+ for each category with log2FC >1.5/<-1.5
If the mean of the controls are differing from the mean of the treatments by >=1.5, you can easily get the significant genes (eg. Control is 9, treatment is 11). However that doesn't justify why all genes are expressed at almost same 'high' level.
Is this the full data? If yes, which genome is this as human genome contains >50k (all kinds of) genes
Yes, its the full data. I'm working on yeast