Reproducibility of Methods to Detect Differentially Expressed Genes from Single-Cell RNA Sequencing

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Differential gene expression analysis is crucial for single-cell RNA sequencing (scRNA-seq) studies.
• Existing methods include those for bulk-cell RNA-seq and specialized scRNA-seq approaches.
• Rigorous statistical assessment is vital due to the unique characteristics of scRNA-seq data.

Purpose of the Study:

• To assess the reproducibility of nine different tools for differential expression analysis in scRNA-seq data.
• To compare methods based on their rediscovery rates (RDRs) for top-ranked genes, distinguishing between highly and lowly expressed genes.
• To evaluate method performance across diverse datasets, including real and simulated scRNA-seq data.

Main Methods:

• Compared nine differential expression analysis tools: four scRNA-seq specific, three bulk-cell adapted, and two general statistical tests.
• Evaluated methods using rediscovery rates (RDRs) for top-ranked genes, stratified by expression level (high vs. low).
• Utilized three real and one simulated scRNA-seq datasets for comprehensive performance assessment.

Main Results:

• Widely used methods like edgeR and monocle showed poorer RDR performance, particularly for top-ranked genes.
• For highly expressed genes, many bulk-cell methods performed comparably to scRNA-seq specific methods.
• For lowly expressed genes, performance varied significantly; edgeR and monocle were liberal (poor false positive control), DESeq2 was conservative (low sensitivity).
• BPSC, Limma, DEsingle, MAST, t-test, and Wilcoxon exhibited similar performances on real datasets.

Conclusions:

• The scRNA-seq specific method BPSC performed well compared to other evaluated tools, especially with sufficient cell counts.
• Method selection for differential expression analysis in scRNA-seq requires careful consideration of gene expression levels and dataset characteristics.
• Some commonly applied bulk-cell methods may not be optimal for scRNA-seq, particularly for detecting lowly expressed differentially expressed genes.