Census cell similarity search: experimental Python API for searching given AnnData #1114
Assignees
Labels
P0
Priority 0 - Critical, fix ASAP!
Comments
This was referenced Apr 28, 2024
|
Per 4/29 discussion: For now (i) the API assumes the given AnnData will include a layer with suitable embeddings and (ii) we'll informally provide a notebook/docker showing how to do the forward pass to add them. To be revisited in H2. |
mlin
added a commit
that referenced
this issue
Aug 8, 2024
Adds two new functions to `cellxgene_census.experimental`: 1. `find_nearest_obs` uses TileDB-Vector-Search indexes of Census embeddings to find nearest neighbors of given embedding vectors (in an AnnData obsm layer). #1114 2. `predict_obs_metadata` uses the nearest neighbors to predict metadata attributes like cell_type and tissue_general for the query cells. Naive implementation is just a starting point to start experimenting with. #1115
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Having built the vector indexes on Census embeddings (#694 #1113), develop the Python API inside
cellxgene_census.experimentalto input an AnnData and identify the most-similar Census cells.This necessitates running a forward pass of the embedding model (starting with scVI) on the given AnnData. In case that causes a lot of complications, we can create an initial demo that searches existing Census cells against the index and break out the forward passes into a separate issue. It might even end up involving a docker image or web service of some sort.
The text was updated successfully, but these errors were encountered: