Hello,
I wanted to analyze and visualize hierarchical clustering (H cluster) and PCA with a subset of genes in my RNASeq data. For instance, I'd like do clustering analyses only with the differentially expressed genes (e.g. pads <0.05) from DESeq2 analysis. (1) Can I take the VST transformed values of the genes of interest from a dataset having all genes that are used for differentially expressed gene analysis and transformed to do the clustering analyses? (2) Or, should I first get the raw counts of a subset genes of interest and then calculate VST transformed values for this subset only? I guess the first approach makes more sense to me, but I wanted to know if this is a valid way to do the analyses.
Thanks for any insight you can share!
Can I take the VST transformed values of the genes of interest from a dataset having all genes that are used for differentially expressed gene analysis and transformed to do the clustering analyses?
Yes, but for a heatmap / hclust one usually transformes these first to the Z-scale, see:
Or, should I first get the raw counts of a subset genes of interest and then calculate VST transformed values for this subset only? I guess the first approach makes more sense to me, but I wanted to know if this is a valid way to do the analyses.
No, absolutely not. The vst makes use of the information of genes with many counts to stabilize the variance for those with low counts. Use it on the full count matrix, then subset for the genes you want to cluster.
For PCA you can use the vst directly, for hclut/heatmap transform to Z first.
Thank you so much for the answer! Just to make sure, for hclust/heatmap, I should normalize VST to Z-score, right? But, could you explain a bit more on why transformation into the z-scale is necessary for hclust/heatmap but not PCA?
Yes, that is correct, see here why: Scaling RNA-Seq data before clustering?
Thank you!
Log in to answer this question.
