PCA plot (Suggestions Needed)

Question

[Aaliya]

Hello, I have 8 samples (4 WT and 4 mutants). I have performed bulk RNA seq on these (Kallisto, and then DESeq2) I am attaching my PCA plot, I was expecting the wild type to cluster together and all the mutants to be together, but this was not the result. Does it indicate that my samples have a lot of variance?

Can anyone suggest to me any other method/plot that can represent my WT vs mutant samples?

I would really appreciate it.

[jkim]

How much variance were explained by PC1 and PC2? It doesn't look like that the largest source of variation came from the genotype difference.

[swbarnes2]

Cross posted here (including the code used to make up the counts):

https://support.bioconductor.org/p/9157911/#9157959

Answer 1

[Jeremy]

Right now, the WT samples are clustering together, but you have one mutant sample off by itself, which could be an outlier. How did you scale the data? If you used Z-scores, I would suggest trying center log ratios.

Besides PCA, you could also try hierarchical clustering.

[Mensur Dlakic]

Everything that Jeremy said, plus that there may be no data manipulation that will give you the clusters you expected. Sometimes there are outliers in the data, and that won't change no matter how much we torture the data.

[Jeremy]

You could also try kernel PCA, which is a non-linear form of PCA, although this will be more time-consuming.

[swbarnes2]

Even if this was real data, the tiny numbers for the PC1 and PC2 indicate that the differences between the samples are all absolutely tiny.

[Aaliya]

I have actually used DESeq2 data.

[swbarnes2]

The code you already posted in the bioconductor board had simulated data, and the PCA image is the same.