Hello Everybody, first I created a PCA of unnormalized RNA-Seq count data. From that result I used the top n genes with the highest absolute loading value and created a heatmap of using their count data. I know the advised way is to normalize the data (accounting for sequencing depth, ..) first and then maybe use the rlog or vst methods from DESeq2 to stabilize the variances.

But since the different samples cluster really good together - as good as after e.g. rlog transformation and the resulted top n genes represent the biological meaning actually really good - is it still a no go?

My gut and my logic say yes. But I was thinking what you guys might think. Thank you in advance and all the best

We discuss in the workflow why variance stabilization is a good idea, so you can just read that (linked from vignette abstract) on my opinion.