Specify the Census search space with an `obs_value_filter` parameter in `find_nearest_obs`

[pablo-gar]

Description

Currently find_nearest_obs finds the closest Census cells to user's data. It would be a great addition to have the ability for the search to be constrained to a subset of user-defined Census cells instead of all Census cells.

To provide consistency with other census API, a simple parameter obs_value_filter (as used in the get_anndataAPI) can be added to limit the search to Census cells meeting the filter criteria.

Context

The find_nearest_obs functionality is great and working as intended, however many times I can be interested on searching the most similar cells against a specific subset of Census defined by a biological context, specially when I know my query cells are from the same/similar biological context.

Impact

This limits the ability to utilize the find_nearest_obs to its full extent in its current form.

Ideal behavior

A parameter obs_value_filter in find_nearest_obs to limit the search to Census cells meeting the filter criteria.

[ivirshup]

Hmm, will have to check with @cathystoli to see if we're still accepting feature requests from you 😉

---

I think this makes a ton of sense as a feature. I've noticed that we probably want to filter out all is_primary==False cells from queries, since you just end up getting a bunch of cells with the exact same embedding.

I've consulted with the TileDB folks and I think this should be quite doable.

• This notebook demonstrates the metadata_df_filter_fn which basically does exactly what you want
• Ideally we would be able to use the same query string you are suggesting, but we need to figure out how to wrap that as a function which can be applied per row.

cc: @mlin