Saving setup_reference output as h5ad file

[robinfallegger]

Hi @earmingol !

It's awesome to have a fast implementation of DOT in python!

I'm running the deconvolution in parallel across many samples, and needed a way to save the output of setup_reference across different runs (and projects!). While building that I ended up with a small bundle of related changes:

• A dot.io module with save_reference/load_reference (plus the underlying dict↔AnnData conversion helpers), so setup_reference()'s output can be written to .h5ad and reloaded.
• DOT() now also accepts that AnnData directly for ref, not just the dict
• Along the way I reorganized the public API into dot.pp/dot.tl/dot.pl namespaces, matching the scanpy/scverse convention (e.g. dot.pp.setup_reference, dot.tl.DOT). The old flat imports (from dotpy import DOT) still work too — no breaking changes here, just a DeprecationWarning nudging toward the namespaced form

Let me know if this sounds useful and I'll open a PR. I am also happy to adjust direction on any of it first.

[earmingol]

Hi @robinfallegger !

Have you tried the CLI? that one allows processing multiple samples simultaneously. However, I think you are considering parallelization, right? The CLI only iterates across samples.

This makes sense to me, the only issue is about keeping parameters consistent, so probably I would add a new function that wraps your preloaded AnnData as a dict matching the DOT input. Maybe you could do this with the load_reference function?

If you want, you can open a PR and we can start iterating from there.

I get the standard of pp, tl, etc, but not sure if that would be pertinent here as we are not dealing with many functions.