I have a query as to why it is necessary to both log transform, and center and scale, scRNA-seq data prior to performing PCA. I thought the purpose of both of these steps was similar - to ensure that genes of different expression levels can contribute similarly to the PCA, rather than highly expressed and thus high variance genes dominating.
- Log transformation reduces heteroskedasticity reducing the association between the mean and variance
- Scaling puts the expression values of all genes on the same scale
Do these not do similar things?