scvi-hub-models

To load pretrained scvi-tools models for query-to-reference mapping, you use the scvi-tools library to download models from the Hugging Face Hub, enabling projection of new single-cell data onto established atlas references.

Yes, you can transfer cell type labels between datasets using scANVI. This Skill supports loading pretrained scANVI models from the Hugging Face Hub to map labels from a reference dataset onto a new query dataset.

For multiomics integration, pretrained TOTALVI and MultiVI models are available. These models support joint embedding analysis across RNA, protein, and ATAC modalities when loaded from the Hugging Face Hub.

scArches works with Hugging Face Hub models by using pretrained scvi-tools models as references. It applies query-to-reference mapping to integrate new single-cell datasets onto existing joint embeddings without full retraining.

Yes, you need online access to the Hugging Face Hub and appropriate permissions to download models. You also need the scvi-tools library installed to perform loading, minified data handling, and embedding analyses.

Use MrVI for multi-sample single-cell analysis when you need to model sample-specific effects. This Skill supports loading pretrained MrVI models to analyze variations across multiple samples within a single dataset.