Bias, robustness and scalability in single-cell differential expression analysis.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) enables gene expression analysis at an unprecedented resolution.
• Numerous computational methods exist for identifying differentially expressed genes from scRNA-seq data, leading to challenges in method selection and result interpretation.

Purpose of the Study:

• To comprehensively evaluate the performance of various differential gene expression analysis methods for scRNA-seq data.
• To compare methods specifically designed for scRNA-seq against those adapted from bulk RNA sequencing.
• To introduce a curated repository of scRNA-seq datasets to facilitate reproducible method evaluation.

Main Methods:

• Systematic evaluation of 36 distinct gene expression analysis algorithms.
• Utilized both experimental and synthetic scRNA-seq datasets for rigorous testing.
• Assessed the impact of prefiltering low-expression genes on method performance.

Main Results:

• Significant variability was observed in the number and identity of differentially expressed genes identified by different methods.
• Prefiltering strategies notably influenced results, particularly for bulk RNA-seq methods applied to scRNA-seq data.
• No consistent performance advantage was found for scRNA-seq-specific methods over well-established bulk RNA-seq approaches.

Conclusions:

• The choice of differential gene expression analysis method significantly impacts scRNA-seq results.
• Bulk RNA-seq analysis methods can be viable alternatives for scRNA-seq differential expression analysis.
• The 'conquer' repository offers standardized datasets to aid future method development and comparative studies.