Using Marimo with scverse mutable objects

[AlbertoFabbri93]

Hi all,

I’ve been experimenting with Marimo recently and appreciate its reactive, immutable execution model. However, I’m running into some friction when trying to use it together with mutable objects from the scverse ecosystem like spatialdata, anndata and mudata.

A lot of typical workflows in scverse rely on in-place mutations (e.g., modifying .obs , .var , or layers directly), which doesn’t seem to align well with Marimo’s approach. In particular, these in-place updates can interfere with change tracking and reactivity.

I do not have much experience with data analysis and pyhton and I would appreciate some advice. For example:

• Do you avoid in-place operations and return modified copies instead?
• Do you structure transformations differently (e.g., fewer, more consolidated steps)?
• Or does this combination end up being more cumbersome than it’s worth?

I’d be interested to hear any experiences, recommendations, or design patterns that make this setup more ergonomic.

Thanks!