Alternating EM algorithm for a bilinear model in isoform quantification from RNA-seq data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• RNA sequencing (RNA-seq) data analysis for isoform expression relies on assumptions that often fail, leading to biased gene expression estimates.
• Existing bias correction methods are limited as they cannot address all known and unknown biases and are applied sample-by-sample.

Purpose of the Study:

• To develop a novel, robust statistical method for accurate isoform-level gene expression estimation from RNA-seq data.
• To introduce a method that automatically corrects for unknown biases in multi-sample RNA-seq analyses.

Main Methods:

• Developed XAEM, a novel method employing a bilinear model (XAEM) for joint estimation of gene expression and bias parameters from multi-sample RNA-seq data.
• Utilized an alternating expectation-maximization (AEM) algorithm for simultaneous estimation of model components.
• Incorporated quasi-mapping for efficient read alignment to ensure computational speed.

Main Results:

• XAEM demonstrates superior accuracy in estimating expression for multi-isoform genes on simulated datasets compared to existing methods.
• Applied to single-cell RNA-seq data, XAEM significantly improved rediscovery rates in differential expression analysis.
• XAEM effectively performs empirical correction of potentially unknown biases, outperforming current advanced methods.

Conclusions:

• XAEM provides a more flexible and robust approach to isoform expression quantification, addressing limitations of current RNA-seq analysis methods.
• The method's ability to auto-correct biases and its high accuracy make it a valuable tool for genomic research, particularly in single-cell studies.