MultiVI

MultiVI generates a joint probabilistic latent representation, implemented in the scvi-tools framework, to integrate single-cell RNA sequencing (scRNA-seq) and assays for transposase-accessible chromatin using sequencing (ATAC-seq) data for studying transcriptional and chromatin accessibility relationships.

Key Features:

• Integration of Multiomic Data: Integrates paired or unpaired scRNA-seq and ATAC-seq datasets to produce a unified representation linking gene expression and chromatin accessibility.
• Handling Missing Modalities: Imputes missing modality measurements so the joint representation remains informative for cells lacking complete multiomic profiles.
• Batch Effect Correction: Models and corrects batch effects at both cell- and sample-levels to mitigate technical confounding across experiments.
• Single Modality Dataset Integration: Incorporates single-modality datasets into the joint latent space to include datasets that only provide one assay.

Scientific Applications:

• Characterizing cellular heterogeneity: Enables joint analysis of expression and chromatin accessibility to resolve heterogeneous cell states at single-cell resolution.
• Regulatory mechanism discovery: Facilitates linking regulatory chromatin features to transcriptional programs to identify putative regulatory relationships.
• Cell differentiation and trajectory analysis: Supports studies of lineage progression by combining chromatin and transcriptional signals to resolve developmental trajectories.
• Disease progression and therapeutic response studies: Allows comparison of multiomic profiles to investigate disease-associated regulatory changes and responses to interventions.

Methodology:

Constructs a probabilistic model that learns a joint latent space by aligning transcriptional and chromatin accessibility data while accounting for batch effects; the integrated representation supports downstream analyses including clustering, differential expression analysis, and trajectory inference.