Beta-Poisson model for single-cell RNA-seq data analyses.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA-sequencing (scRNA-seq) enables gene expression analysis at the individual cell level.
• scRNA-seq data often exhibits bimodality in gene expression distributions, a characteristic not captured by traditional bulk RNA-seq models.
• Existing models like gamma-Poisson struggle with the high proportion of non-expressing cells common in scRNA-seq.

Purpose of the Study:

• To develop a statistical model capable of handling the unique bimodality of single-cell gene expression data.
• To integrate this model into a generalized linear model framework for robust differential expression analysis in scRNA-seq.
• To introduce the BPSC analytical procedure and R package for scRNA-seq data analysis.

Main Methods:

• Introduction of a beta-Poisson mixture model to address gene expression bimodality.
• Integration of the beta-Poisson model within a generalized linear model framework for differential expression analysis.
• Development of the BPSC (Beta-Poisson Single-Cell) analytical procedure and corresponding R package.

Main Results:

• The beta-Poisson model effectively characterizes gene expression for approximately 90% of transcripts in scRNA-seq datasets.
• BPSC demonstrates superior model-fitting compared to the standard gamma-Poisson model in over 80% of transcripts.
• BPSC shows competitive performance against established methods (edgeR, scde, MAST) in differential expression analyses of both simulated and real scRNA-seq data.

Conclusions:

• The beta-Poisson mixture model provides a statistically sound approach for analyzing bimodal gene expression in scRNA-seq data.
• BPSC offers an effective and accurate method for differential expression analysis, outperforming existing tools.
• The BPSC R package facilitates the application of this advanced methodology for scRNA-seq data analysis.