alterlab-scvi-tools

Batch correction for single-cell RNA-seq data is performed by training variational autoencenders on raw counts with batch covariates to generate integrated latent representations for downstream analysis.

Differential expression analysis on multimodal single-cell data is best run using a Bayesian probabilistic framework, which leverages trained latent representations to account for technical noise and batch effects.

Yes, scvi-tools supports spatial transcriptomics and ATAC-seq integration, enabling end-to-end probabilistic modeling across RNA-seq, ATAC, and multimodal assays within a unified framework.

To set up a single-cell model for multimodal analysis, you need an AnnData object containing raw counts and explicit covariates for batch correction and other biological factors.

Yes, scvi-tools supports cross-modality imputation, using trained probabilistic models to predict missing modalities and perform uncertainty-aware differential testing across batches.

Single-cell batch correction models require raw counts because the underlying probabilistic VAE framework models count distributions directly, ensuring accurate likelihood estimations and biological variance capture.