Heavy-tailed prior distributions for sequence count data: removing the noise and preserving large differences

Area of Science:

• Bioinformatics
• Genomics
• Statistical Genetics

Background:

• RNA-sequencing (RNA-seq) is crucial for differential gene expression analysis.
• Accurate estimation of effect size, specifically logarithmic fold change (LFC), is challenging due to biological and technical variability.
• Low or variable read counts lead to high variance in LFC estimates, compromising gene ranking and biological interpretation.

Purpose of the Study:

• To develop a robust method for estimating gene expression effect sizes in RNA-seq data.
• To address limitations of existing methods like filtering and pseudocounts.
• To improve the accuracy and reliability of differential expression analysis.

Main Methods:

• Proposed a novel approach using a heavy-tailed Cauchy prior distribution for effect sizes.
• Developed the Approximate Posterior Estimation for generalized linear model (apeglm) method.
• Integrated apeglm as an R/Bioconductor package and within DESeq2 software.

Main Results:

• Apeglm effectively reduces bias and variance in LFC estimates, outperforming existing shrinkage estimators.
• The method avoids the need for arbitrary filtering thresholds or dataset-specific pseudocounts.
• Improved statistical inference for genes with limited information, leading to more reliable differential expression detection.

Conclusions:

• Apeglm offers a superior approach to estimating gene expression effect sizes in RNA-seq.
• The method enhances the accuracy of differential expression analysis, particularly in challenging datasets.
• Apeglm provides a valuable tool for researchers in genomics and bioinformatics.