Modeling expression ranks for noise-tolerant differential expression analysis of scRNA-seq data

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell transcriptomics provides high-resolution insights into cellular heterogeneity.
• Accurate modeling of gene expression is crucial for identifying tissue-specific patterns.
• Existing methods often struggle with high dropout rates common in single-cell data.

Purpose of the Study:

• To explore modeling gene expression ranks as an alternative to expression estimates.
• To develop and evaluate a novel differential expression test for single-cell data.
• To assess the performance of the discrete generalized beta distribution (DGBD) for this purpose.

Main Methods:

• Utilized the discrete generalized beta distribution (DGBD) to model gene expression ranks.
• Devised a Wald-type test for comparing gene expression between two single-cell groups.
• Developed the ROSeq R package for method dissemination.

Main Results:

• The proposed ROSeq method demonstrated a good balance between Type I and Type II errors.
• ROSeq showed exceptional robustness to expression noise.
• The method exhibited rapid scalability with increasing sample sizes.

Conclusions:

• ROSeq provides a robust and scalable solution for differential gene expression analysis in single-cell transcriptomics.
• Modeling gene expression ranks offers advantages over traditional expression-based approaches.
• The ROSeq R package is available on Bioconductor for broader adoption.