Error estimates for the analysis of differential expression from RNA-seq count data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA-sequencing (RNA-seq) analysis for differential gene expression relies on statistical algorithms.
• Overdispersed count data in RNA-seq presents challenges, leading to inaccurate false discovery rate (FDR) estimations and reduced statistical power.
• Non-uniform p-value distributions under the null hypothesis are a common issue in existing methods.

Purpose of the Study:

• To evaluate the accuracy of FDR estimation across various R packages for RNA-sequencing data analysis.
• To compare the performance of established tools like edgeR, DESeq2, and QuasiSeq, along with a novel package, Polyfit.

Main Methods:

• Utilized synthetic and real biological RNA-seq datasets for performance assessment.
• Surveyed R packages including edgeR, DESeq, DESeq2, PoissonSeq, and QuasiSeq.
• Introduced and tested Polyfit, an add-on package designed to improve FDR estimation using the Storey-Tibshirani procedure.

Main Results:

• QuasiSeq (QLSpline implementation) demonstrated the most accurate FDR estimation across all p-values, though with longer computation times, especially with >= 4 biological replicates.
• edgeR and DESeq2 were identified as the next best performing packages.
• The Polyfit extension enhanced DESeq performance to levels comparable with edgeR and DESeq2 when >= 6 biological replicates were used.

Conclusions:

• QuasiSeq offers superior FDR accuracy for differential gene expression analysis in RNA-seq, particularly with sufficient biological replicates.
• edgeR and DESeq2 provide robust performance for FDR estimation.
• Polyfit presents a viable option for improving DESeq's accuracy in studies with a higher number of replicates.