scvi-tools

Why we included this project

Single-cell experiments generate high-dimensional, noisy data, and scvi-tools makes probabilistic modeling of it practical without hand-rolling your own variational models. The package ships prebuilt models for the tasks that come up constantly in single-cell work: dimensionality reduction, batch integration across datasets, automated cell-type annotation, factor analysis, doublet detection, and spatial deconvolution. They all share a consistent high-level API that slots into the Scanpy ecosystem, so anyone used to standard scRNA-seq analysis will feel at home. The underlying stack is PyTorch Lightning and Pyro, which brings GPU acceleration and standard save/load behavior along with it, and the same building blocks can be extended when an experiment does not fit a stock model. For Python-savvy teams that would rather have a coherent foundation than wire together scattered scripts, it is a practical pick.