SCTransform workflow steps on multi-sample datasets

[MarcElosua]

Hello! Thank you for developing this package and for all the documentation available.

TLDR: Recommended workflow for multi-sample datasets SCT normalization to merging and PCA.

I have a question about the usage of SCTransform in the context of multiple-sample datasets where we don't want to integrate directly. As I understand from reading these issues: issue1, issue2, and issue3 I understand that when I have a multisample dataset I should be running SCTransform by each 10X experiment to correct for experiment specific technical noise. This also helps clear memory requirements with very large datasets 500K+ cells. Once merged, we can obtain the highly variable genes from the dataset like this:

pancreas.list <- lapply(X = pancreas.list, FUN = SCTransform)
features <- SelectIntegrationFeatures(pancreas.list)
VariableFeatures(merged_object) <- features

Next I am interested in computing PCA on these samples to assess if integration is even necessary and I was wondering if I should do ScaleData on the HVG or not?

Thank you very much for any help!