Deep generative modeling for single-cell transcriptomics.

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Single-cell RNA sequencing (scRNA-seq) offers insights into cellular heterogeneity.
• Technical noise and biases in scRNA-seq data complicate downstream analysis and interpretation.
• Accurate modeling of uncertainty is crucial for reliable single-cell data analysis.

Purpose of the Study:

• Introduce single-cell variational inference (scVI), a scalable probabilistic framework.
• Provide a ready-to-use tool for the analysis of gene expression in single cells.
• Address technical noise, batch effects, and limited sensitivity in scRNA-seq data.

Main Methods:

• Developed scVI, a framework leveraging deep neural networks and stochastic optimization.
• scVI aggregates information across cells and genes to approximate expression distributions.
• The model accounts for technical noise, batch effects, and limited sensitivity.

Main Results:

• Applied scVI to fundamental single-cell analysis tasks: batch correction, visualization, clustering, and differential expression.
• Demonstrated high accuracy across all tested analysis tasks.
• scVI provides a robust probabilistic representation of gene expression data.

Conclusions:

• scVI is an effective and scalable framework for single-cell gene expression analysis.
• The probabilistic approach improves accuracy in downstream analyses.
• scVI facilitates a deeper understanding of biological diversity from scRNA-seq data.