[Question] How to extract all gene embeddings from one foundational model ?

[LPioL]

Hi,

It is not clear to me how to extract all gene embeddings from one of the models. I tried this code but can only extract from one specific tissue and I am not sure my code is right and do what I want. Any help will be super appreciated. My code:

emb_names = ["scgpt"]
census_version = "2023-12-15"

with cellxgene_census.open_soma(census_version=census_version) as census:
    adata = cellxgene_census.get_anndata(
        census,
        organism="homo_sapiens",
        measurement_name="RNA",
        obs_value_filter="tissue == 'tongue'",
        obs_embeddings=emb_names,
    )

# Get gene names
gene_names = adata.var.feature_name.tolist()  # List of gene names

# Convert sparse matrix to dense format
gene_embeddings_dense = adata.X.toarray()

# Map genes to their embeddings
gene_to_embedding_dict = {
    gene: gene_embeddings_dense[:, idx].tolist()  # Extract all embeddings for the gene
    for idx, gene in enumerate(gene_names)
}

[johnkerl]

cc @ivirshup