Modeling Exon-Specific Bias Distribution Improves the Analysis of RNA-Seq Data

Area of Science:

• Transcriptomics
• Bioinformatics
• Computational Biology

Background:

• RNA-sequencing (RNA-seq) is crucial for gene and transcript expression quantification.
• Key challenges include read mapping ambiguity and overdispersion in read distribution.
• Existing methods often use Poisson models, which struggle with count splitting and overdispersion.

Purpose of the Study:

• To develop a novel method, PGseq, to accurately quantify gene and transcript expression from RNA-seq data.
• To address challenges of read mapping ambiguity and overdispersion in read counts.
• To improve accuracy in transcriptome and differential expression analyses, particularly for low-expression genes.

Main Methods:

• Introduced Gamma-distributed latent variables to model exon-specific read sequencing preferences.
• Integrated these variables into a Poisson model's rate parameter to account for read count overdispersion.
• Employed maximum likelihood estimation with integrated Gamma priors for model tractability.

Main Results:

• PGseq demonstrated competitive performance against popular methods in accuracy for gene and transcript expression quantification.
• The method showed advantages in downstream differential expression analysis.
• PGseq particularly excelled in analyzing low-expression genes, outperforming alternatives.

Conclusions:

• PGseq effectively models exon-specific read preferences and overdispersion in RNA-seq data.
• The approach provides accurate gene and transcript expression quantification and enhances differential expression analysis.
• PGseq offers a valuable tool for transcriptome studies, especially when analyzing low-expression transcripts.