salient space and downstrema for normal cells?

[jjuhyunkim]

Hi,

I have a question about the downstream analysis using results from contrastive VI. I believed that all the values in the salient latent space were set to zero. However, when I generated a UMAP plot with all the cells and samples, including normal and case samples, I was able to observe a UMPAP plot that looks very reasonable. In this plot, normal cells clustered closely rather than with case cells.

So my question is, after model training, is it impossible or not informative to conduct downstream analyses such as connectivity, dimension reduction, DEG or others? If so, is it correct that the only way to use the salient space and other functions related to salient results is for un-normal cells by extracting them from the original data?

It would be great if you could double-check my thoughts or correct me if I have misunderstood some points.

Thank you!