Differential expression of single-cell RNA-seq data using Tweedie models.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) data analysis is sensitive to normalization methods and experimental platforms.
• Over 100 differential expression (DE) tools exist, each with unique assumptions for modeling scRNA-seq data.
• Technological variability across platforms complicates accurate DE analysis.

Purpose of the Study:

• To develop a computational method that models technological variability in cross-platform scRNA-seq data.
• To improve the identification of differentially expressed genes in scRNA-seq studies.
• To provide a robust tool for analyzing diverse scRNA-seq datasets.

Main Methods:

• Proposed Tweedie generalized linear models to capture diverse scRNA-seq expression profiles.
• Introduced a zero-inflated Tweedie model for data with excessive zero counts.
• Conducted a systematic benchmark evaluation of over 10 DE methods using synthetic and real scRNA-seq datasets.

Main Results:

• Tweedieverse effectively models platform- and gene-specific statistical properties like heavy tails and sparsity.
• The proposed zero-inflated Tweedie model addresses zero-inflated scRNA-seq data.
• Tweedieverse demonstrated superior performance compared to state-of-the-art DE methods in benchmark evaluations.

Conclusions:

• Tweedieverse offers enhanced statistical power and better false discovery rate control for scRNA-seq DE analysis.
• The method performs robustly across different experimental platforms (plate- and droplet-based).
• An open-source R/Bioconductor package is available for public use.