Swish

Swish performs nonparametric differential expression analysis of RNA-seq count data by incorporating inferential replicate counts to account for quantification uncertainty and control false discovery rate at gene and transcript levels.

Key Features:

• Nonparametric approach: Employs a nonparametric model that does not assume a specific distribution for count data, enabling flexible application across data types without reformulating statistical models.
• Incorporation of inferential uncertainty: Integrates inferential replicate counts to account for uncertainty in abundance estimates during quantification, improving robustness of differential expression calls.
• Extension of SAMseq: Extends the SAMseq methodology by incorporating inferential replicate counts into the nonparametric testing framework.
• False discovery rate control: Provides improved control of the false discovery rate (FDR), especially for transcripts with high inferential uncertainty.
• Comparative performance: Demonstrates superior performance relative to methods including the Wilcoxon test, notably for single-cell RNA-seq and for assessing differential expression between cell sub-populations.

Scientific Applications:

• Single-cell RNA-seq analysis: Assessing differential expression between sub-populations of cells while accounting for cell-to-cell inferential uncertainty.
• Comparative genomics studies: Performing differential expression analyses across species or conditions without requiring distributional model reformulation.
• Transcriptome-wide studies: Large-scale gene- and transcript-level differential expression testing with enhanced FDR control to reduce false positives.

Methodology:

Swish applies a nonparametric extension of SAMseq that integrates inferential replicate counts from RNA-seq quantification to perform differential expression testing while accounting for inferential uncertainty.