scvi-tools

Batch correction on scRNA-seq data is performed using variational autoencoders that model batch covariates, integrating multi-batch experiments to output a corrected latent representation.

Probabilistic modeling supports multimodal single-cell data integration by applying variational inference to jointly analyze scRNA-seq and scATAC-seq datasets, yielding cross-modality latent representations and denoised expressions.

scvi-tools models raw count data using variational autoencoders with Negative Binomial (NB), Zero-Inflated Negative Binomial (ZINB), or Poisson likelihoods to capture single-cell omics noise and dropout.

To get a latent representation for spatial transcriptomics, the Skill applies probabilistic modeling to raw counts, performing dimensionality reduction while handling technical noise and batch effects.

Yes, scvi-tools supports differential expression analysis across modalities by leveraging denoised expression outputs from the trained variational models, enabling cross-modality biological insights.

The best way to integrate multi-batch single-cell experiments is using probabilistic models that handle batch covariates directly, correcting technical variations while preserving biological signals in latent space.