Isolating salient variations of interest in single-cell data with contrastiveVI

Area of Science:

• Computational Biology
• Genomics
• Immunology

Background:

• Single-cell RNA sequencing (scRNA-seq) is crucial for studying cellular state changes under different conditions.
• Existing computational models struggle to distinguish variations unique to treatment from those common to control groups.
• Understanding treatment response heterogeneity requires deconvolving these distinct variation types.

Purpose of the Study:

• To introduce contrastive variational inference (contrastiveVI), a novel framework for analyzing scRNA-seq data.
• To enable the separation of shared and treatment-specific latent variables in single-cell datasets.
• To enhance the analysis of treatment effects and cellular heterogeneity.

Main Methods:

• Developed contrastive variational inference (contrastiveVI), a probabilistic framework.
• Applied contrastiveVI to analyze three treatment-control scRNA-seq datasets.
• Extended contrastiveVI for joint analysis of transcriptome and surface protein data.

Main Results:

• ContrastiveVI successfully deconvolves shared and treatment-specific variations in scRNA-seq data.
• The framework demonstrated strong agreement with known biological ground truths across multiple datasets.
• ContrastiveVI identified subtle biological phenomena often missed by standard analysis workflows.

Conclusions:

• ContrastiveVI provides a powerful new tool for dissecting cellular heterogeneity in response to treatments.
• The framework improves visualization, clustering, and differential expression analysis in scRNA-seq studies.
• Generalization to multi-omics data (transcriptome and surface proteins) expands its utility.