scvi-tools

You can integrate single-cell RNA-seq datasets using the scVI model for batch correction, which learns latent representations that remove technical batch effects while preserving biological variability. This Skill provides guidance on model selection, data preparation, training, and validation procedures.

To analyze CITE-seq and multiome data together, use totalVI and MultiVI models respectively to jointly analyze paired molecular modalities. This Skill helps you select the correct probabilistic model, prepare paired data, and interpret integrated multiomic results.

Deep learning improves single-cell genomics analysis by applying probabilistic models to learn non-linear latent representations of complex data distributions. This approach enables advanced tasks like reference mapping, spatial deconvolution, and RNA velocity without designing models from scratch.

Yes, you can map new single-cell data to an existing annotated reference using scArches for label transfer and reference mapping. This Skill supports workflows for projecting new queries onto trained latent spaces to annotate new datasets efficiently.

For spatial transcriptomics deconvolution, choose the DestVI model to estimate cell type proportions and cell-type-specific gene expression within spatial spots. This Skill provides model selection guidance to match your biological data type to the correct probabilistic architecture.

No, you do not need to design your own probabilistic models for RNA velocity analysis; you can use the veloVI model directly. This Skill enables researchers to apply advanced deep learning methods to single-cell genomics without building probabilistic architectures from scratch.