BBmix: a Bayesian beta-binomial mixture model for accurate genotyping from RNA-sequencing.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Existing genotype calling pipelines for RNA sequencing (RNA-Seq) adapt DNA methods, failing to account for RNA-Seq specific biases like allele-specific expression (ASE).

Purpose of the Study:

• To introduce BBmix, a novel Bayesian beta-binomial mixture model designed to accurately call genotypes from RNA-Seq data by modeling RNA-Seq specific biases.
• To improve the accuracy of heterozygous calls and reduce false positive rates in downstream applications sensitive to genotyping errors.

Main Methods:

• Developed BBmix, a Bayesian beta-binomial mixture model that learns genotype-specific read count distributions.
• Probabilistically calls genotypes using learned parameters, outperforming existing methods on diverse datasets.

Main Results:

• BBmix demonstrated superior performance compared to existing methods, achieving up to a 1.4% increase in heterozygous call accuracy.
• The model showed parameter transferability across datasets, enabling efficient genotype calling for numerous samples after a single learning run (<1 hour).

Conclusions:

• BBmix offers a more accurate and efficient approach to genotype calling from RNA-Seq data.
• Its improved accuracy, particularly for heterozygous calls, has significant implications for applications like ASE analysis, reducing false positives.