Improving SCVI for low-count cells through self-supervised augmentation

[chelseabright96]

Hi,

A paper has recently been written about improving scVI for low count cells using binomial thinning and a cross-correlation loss. It looks like this extension of scVI improves on the integration of low count cells based on the paper results. I thought this may be of interest for you and that you may consider incorporating this into scvi-tools at some point.

Paper link: https://www.biorxiv.org/content/10.64898/2026.02.11.705441v1.full.pdf

[ori-kron-wis]

Thank you @chelseabright96 , we are well aware of it and have been collaborating with Valentine for a long time, so we might as well implement it.