Reproducing scVI result in Seurat v5 vignette

[shwong-tw]

Dear developers,

I was going through this vignette Integrative analysis in Seurat v5 and would like to reproduce the figures, mainly to ensure that different integration methods were all configured properly in my pc.

When it comes to comparing CCA and scVI results side by side in this section, the first figure in the vignette suggested that both CCA and scVI managed to have cells from different technologies mixed together rather nicely.
However in my hand scVI result did not show similar performance.

The vignette's session info did not contain information about conda environment (where scVI was installed),
therefore I ran the vignette under two different conda environment (scVI version 1.0.4 vs 1.2.0), with the results shown below:
(A) the figure in the origial vignette

(B) running vignette under conda installed with scVI version 1.0.4

(C) (B) running vignette under conda installed with scVI version 1.2.0

It seems to me both (B) scVI v1.0.4 and (C) scVI 1.2.0 did not have all technologies integrated very well as in the vignette (A), e.g. looking at the reddish cells.

It would be great to hear some feedback from the developers,
whether you'd be able to provide more information about the scVI conda environment used in the vignette,
or whether you (or anyone) would observe the same phenomenon when you re-run this vignette.

Thank you very much!