A New Machine Learning-Based Framework for Mapping Uncertainty Analysis in RNA-Seq Read Alignment and Gene Expression...

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• RNA-Sequencing (RNA-Seq) provides more accurate gene expression estimations than microarrays.
• Multiple-mapping reads (MMRs) pose a significant challenge, affecting gene expression quantification and downstream analyses in plant, animal, and metagenome samples.
• Current RNA-Seq analysis pipelines lack robust methods to evaluate the reliability of gene expression estimations.

Purpose of the Study:

• To develop and validate a machine learning-based tool, GeneQC, for assessing the reliability of gene expression levels derived from RNA-Seq data.
• To address the impact of MMRs on gene expression accuracy and downstream analysis validity.
• To provide researchers with a method to identify reliable gene expression estimations and guide further analysis.

Main Methods:

• Developed GeneQC, a machine learning tool utilizing genomic and transcriptomic features.
• Employed elastic-net regularization and mixture model fitting to quantify mapping uncertainty for each gene.
• Analyzed 95 RNA-Seq datasets across seven plant and animal species, identifying an average of 22% MMRs.

Main Results:

• GeneQC accurately estimates the reliability of gene expression levels.
• The study found an average of approximately 22% MMRs across diverse species.
• Plant samples exhibited high mapping uncertainty, while animal samples showed limited but severe uncertainty.

Conclusions:

• GeneQC enhances the reliability of RNA-Seq-based gene expression analysis by quantifying mapping uncertainty.
• The tool empowers researchers to discern trustworthy expression data and improve downstream analyses.
• GeneQC facilitates the identification of genes with low reliability, prompting further investigation and more accurate expression estimation methods.