Scvi embedding from Census API by CellxGene

Help scvi-tools
1

Hi there,

I’m quite new to scvi. So please excuse if there is some notion mistake.

What I try to achieve is accessing the full dataset’s scvi embeddings from the model page of Census by CellxGene platform(s3://cellxgene-contrib-public/models/scvi/2024-07-01/mus_musculus/model.pt)
Then, using this pretrained embeddings (if possible) to predict the cell types of my query dataset.

The given example on the Census is for small slice of the anndata object from Census.
For the full model of the ‘model.pt’ file I could not find a working approach to predict the cell types of my query anndata object (actually, converted Seurat object).

2

Hi, please direct user questions of the CELLxGENE page to the CELLxGENE team. While we help with model training and review their pipeline, it is different to ours at Querying the Human Lung Cell Atlas — scvi-tools (smaller dataset but I would run similar a KNN on the whole data as was done in: https://www.biorxiv.org/content/10.1101/2024.03.01.582887v1 (see reproducibility for running kNN on GPU.