BBmix

BBmix applies a Bayesian beta-binomial mixture model to call genotypes from RNA-Seq data by modeling RNA-Seq-specific biases, including allele-specific expression, to improve genotyping accuracy.

Key Features:

• Bayesian framework: Employs a Bayesian beta-binomial mixture model that learns expected read-count distributions for each genotype and produces probabilistic genotype calls.
• Improved accuracy: Demonstrates improved heterozygous genotype calling, reported as up to a 1.4% increase in accuracy in benchmarking studies.
• Bias modeling: Explicitly models RNA-Seq-specific biases, including allele-specific expression, to reduce false positives in genotype calls.
• Efficiency and transferability: Parameters learned are transferable within datasets, and a single learning run can be completed in under one hour for genotyping multiple samples.
• Integration into pipelines: Can be incorporated into standard genotype-calling pipelines for use with RNA-Seq-derived read counts.

Scientific Applications:

• Allele-specific expression: Enables more accurate inference of allele-specific expression by reducing genotyping errors from RNA-Seq data.
• Genetic association analyses: Provides improved genotype calls from RNA-Seq for downstream genetic association analyses where genotype accuracy affects results.
• RNA-Seq-based genotyping: Supports studies that infer genotypes directly from RNA-Seq data for downstream analyses sensitive to genotyping errors.

Methodology:

Uses a Bayesian beta-binomial mixture model to learn read-count distributions per genotype and make probabilistic genotype calls while explicitly modeling RNA-Seq biases such as allele-specific expression; parameters are transferable across samples and a single training run is reported to take less than one hour.