predictions after scvi integration

[pirlowang]

I have run scvi integration for my 1.5 millions scRNA-seq data from multiple patients. when I run predictions = celltypist.annotate(adata, model = 'Adult_Human_Skin.pkl', majority_voting = True). It warning:
WARNING:celltypist.logger:⚠️ Warning: invalid expression matrix, expect all genes and log1p normalized expression to 10000 counts per cell. The prediction result may not be accurate

Then I run adata = adata.raw.to_adata() and rerun annotation. It goes well but running leiden clustering again. I never get this run done because the kernal is killed even if I set 100G memory.

The scvi command:
adata.layers['raw_counts']=adata.X.copy()
sc.pp.normalize_total(adata,target_sum=1e4)
sc.pp.log1p(adata)
adata.raw = adata
adata.raw = adata
scvi.model.SCVI.setup_anndata(adata, layer="raw_counts", batch_key='donor', categorical_covariate_keys=['platform'])
model = scvi.model.SCVI(adata, n_layers=1, n_latent=6, gene_likelihood="zinb")
model.train()
sc.pp.neighbors(adata, use_rep=SCVI_LATENT_KEY)
sc.tl.leiden(adata)
sc.tl.umap(adata)

[ChuanXu1]

@pirlowang, what's the shape of your raw count matrix? Does it contain only HVGs or all genes?

[pirlowang]

adata has a layer 'raw_counts', and adata.raw.X contains all normalized gene counts. adata.X has only HVGs. here is my HVGs command:
sc.pp.highly_variable_genes(
adata,
flavor="seurat_v3",
n_top_genes=2000,
layer="counts",
batch_key="batch",
subset=True,
)