scvi-tools

Batch effects in single-cell RNA-seq data are corrected using a probabilistic modeling framework that generates batch-corrected latent representations. By configuring batch covariates in AnnData, the model learns and removes technical variation across datasets.

Probabilistic multimodal integration for single-cell omics is a modeling approach that jointly analyzes scRNA-seq, scATAC-seq, CITE-seq, and spatial transcriptomics. It enables cross-modality imputation and unified cell-type annotation by learning shared latent representations.

Differential expression analysis can be performed directly on raw count input. The unified probabilistic framework processes raw single-cell counts to denoise data and compute statistically robust differential expression results across cell populations.

scvi-tools works with AnnData for spatial transcriptomics and CITE-seq datasets. It provides modular model choices across these diverse modalities, handling raw-count input to produce latent representations for downstream dimensionality reduction.

The best way to integrate scATAC-seq and scRNA-seq datasets is using a unified probabilistic framework that performs multimodal integration. This approach aligns cross-modality profiles in a shared latent space for joint analysis and imputation.

GPU-accelerated training is not strictly required but is supported for single-cell dimensionality reduction. It significantly speeds up the training of probabilistic models on large multimodal datasets to obtain latent representations faster.