Scvi Tools

If you're doing single-cell genomics work, this wraps scvi-tools, the PyTorch-based framework for probabilistic models in single-cell analysis. It handles the usual suspects: scRNA-seq batch correction and integration, CITE-seq multimodal data, spatial transcriptomics deconvolution, ATAC-seq. The whole library follows a consistent setup-train-extract pattern with AnnData objects, so it plays nice with scanpy. The real value is in the variational inference approach for things like differential expression and batch effects, where you get actual probabilistic outputs instead of just point estimates. Works with GPU acceleration for large datasets. If you're still wrestling with harmony or Seurat integration issues, the scVI and scANVI models here are worth trying.