Probabilistic models for single-cell omics data

Dimensionality reduction, dataset integration, differential expression, automated annotation. scvi-tools contains models that perform a wide variety of tasks across many omics, all while accounting for the statistical properties of the data.

Each model (e.g., scVI, scANVI, Stereoscope, totalVI) follows the same user interface that couples nicely with Scanpy, Seurat, or Bioconductor workflows. No more searching through source code.

Building novel probabilistic models with scvi-tools is simplified by its object-oriented design and base components powered by PyTorch, PyTorch Lightning, Pyro, and AnnData. No need to write a dataloader or trainer ever again.

scvi-tools models are trained efficiently through minibatching and can naturally be used with a GPU. These models are prepared to scale with growing dataset sizes.