scvi-tools-single-cell

Batch correction for scRNA-seq data integrates multiple experiments while preserving biological variation using probabilistic models like scVI. This approach requires AnnData objects with raw counts to accurately align datasets and remove technical noise.

Semi-supervised cell type annotation assigns labels to unlabeled scRNA-seq cells using a partially labeled reference with scANVI. It leverages probabilistic generative models to transfer learned representations and predict missing cell types robustly.

Multi-omics CITE-seq integration is supported through totalVI, which jointly models RNA and protein expression. This deep generative approach processes raw counts in AnnData objects to extract unified representations of cellular states across modalities.

Adapting pre-trained single-cell models to new datasets is achieved efficiently via transfer learning with scARCHES. This method fine-tunes existing deep generative representations on new scRNA-seq or multi-omics data without requiring full retraining.

Differential expression analysis with uncertainty quantification is performed using deep generative models on single-cell data. By leveraging probabilistic frameworks, it yields statistically grounded comparisons of gene expression between identified cell subtypes.

Raw counts in AnnData objects are strictly required to run deep generative models for single-cell omics analysis. The probabilistic models rely on unnormalized count data to accurately capture technical noise and biological variation during batch correction.