Understanding scVI integration inside R with Seurat v5 & SCTransform

Help scvi-tools
2

Hey @y.grabovska

Thank you for posting here. I think your analysis is correct.

  1. For integration SCVI-Tools need to have the raw counts.
  2. You can also think about the union of HVG’s per group, not only intersection. how much they differ? But, there are method that also compensate for the group size or batch like Scanpy’s highly_variable_genes() with batch_key (Python) or Seurat’s SelectIntegrationFeatures() (R). You can downsample the groups sizes to match each other and repeaet this process several times.
  3. Its ok to manually remove HVG that are known to be in fact, noise.
Can scVI take as input SCT-normalized data?