Has anyone used Marimo together with Scanpy or SpatialData?

I’ve been experimenting with Marimo and like its reactive, immutable execution model, but I’m running into friction when working with Scanpy/SpatialData objects. Many typical workflows rely on in-place mutations which doesn’t seem to fit naturally with Marimo’s approach.For example, operations that modify .obs, .var, or layers in place break change tracking and reactivity.

Has anyone found a good pattern for using these tools together? Do you adapt your workflow (e.g., avoid in-place ops, copy more aggressively, or consolidate transformations into a single cell), or does it end up being more trouble than it’s worth?

Curious to hear real experiences or best practices.