A Poisson reduced-rank regression model for association mapping in sequencing data

Area of Science:

• Genomics
• Computational Biology
• Statistical Genetics

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at the individual cell level.
• Understanding associations between transcriptional activity and cell-specific covariates (e.g., cell type, genotype, health) is crucial.
• Traditional methods struggle with computational cost, ignoring gene correlations, and count-based data assumptions.

Purpose of the Study:

• Develop a reduced-rank regression model for scRNA-seq data.
• Identify low-dimensional linear associations between cell covariates and high-dimensional gene expression.
• Address limitations of traditional methods in handling scRNA-seq data characteristics.

Main Methods:

• Developed a probabilistic, reduced-rank regression model.
• Incorporated a Poisson likelihood to handle count-based scRNA-seq data.
• Validated the model through simulations and application to diverse RNA-seq datasets.

Main Results:

• The model effectively identifies associations between gene expression and cell-specific covariates.
• Demonstrated the model's performance on scRNA-seq, spatial gene expression, and bulk RNA-seq data.
• Leveraged low-dimensional representations to capture transcriptional states.

Conclusions:

• The reduced-rank regression approach provides a powerful statistical framework for scRNA-seq analysis.
• This method effectively links gene expression patterns to cellular and sample-specific factors.
• The approach offers a computationally efficient and statistically robust way to analyze complex transcriptomic data.