scVI corrected dimensions with MiloR

[LinearParadox]

Hi all,

I was wondering whether scVI dimensions can be used with miloR. Specifically, each cell is transformed into a low dimensional latent vector of normal random variables. The issue I see possibly occurring is that this vector only has 10 dimensions, instead of your typical 20-40 for PCs. Using algorithms like Leiden and UMAP, I've been getting pretty good results just based on the 10D scVI output (it is basically a latent embedding and used like PCs) . However, I wasn't quite sure how heavily this would affect miloR specifically.

Publication for reference:
https://www.nature.com/articles/s41592-018-0229-2

[MikeDMorgan]

Hi @LinearParadox - Milo only cares about what happens once you've built the graph - there is no implicit barrier to using an scVI embedding to build your NN-graph - for example see: https://www.nature.com/articles/s41588-023-01523-7

If you're concerned about the small number of dimensions, then I'd recommend running a sensitivity analysis with some positive control data.