Nonparametric expression analysis using inferential replicate counts.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) data analysis faces challenges in identifying differentially expressed genes while managing technical biases.
• Existing statistical methods often use parametric models, but nonparametric approaches may offer better false discovery rate (FDR) control and adaptability.
• Current nonparametric methods for RNA-seq analysis do not fully address inferential uncertainty, potentially inflating FDR, particularly at the transcript level.

Purpose of the Study:

• To propose a novel nonparametric model for RNA-seq differential expression analysis that incorporates inferential uncertainty.
• To extend the SAMseq method to better control FDR by utilizing inferential replicate counts.
• To evaluate the performance of the proposed method, Swish, against existing popular differential expression analysis techniques.

Main Methods:

• Development of Swish, a nonparametric model for differential expression analysis in RNA-seq data.
• Incorporation of inferential replicate counts to account for uncertainty in abundance estimates.
• Comparison of Swish's performance against established methods using simulated and real RNA-seq datasets, including single-cell RNA-seq data.

Main Results:

• Swish demonstrates improved control of the false discovery rate compared to popular methods.
• The proposed method shows particular benefit for transcripts exhibiting high inferential uncertainty.
• Application to a single-cell RNA-seq dataset highlights Swish's utility in identifying differential expression between cell sub-populations.

Conclusions:

• Swish offers a robust nonparametric approach for RNA-seq differential expression analysis, effectively managing inferential uncertainty.
• The method provides enhanced FDR control, making it a valuable tool for transcript-level analysis.
• Swish shows promise for analyzing complex datasets like single-cell RNA-seq, aiding in the identification of biologically relevant gene expression differences.