BIRD

BIRD applies Bayesian hierarchical modeling to identify regulatory variants and quantify allelic effects in high-throughput reporter assays such as STARR-seq and other massively parallel reporter assays (MPRAs).

Key Features:

• Bayesian Hierarchical Model: Integrates prior information and explicitly accounts for variability across experimental replicates to estimate allelic effects.
• Improved Statistical Significance Assessment: Leverages the independent assay of loci in reporter assays to assess significance without confounding from linkage disequilibrium.
• Handling Low Allele Frequencies: Provides enhanced performance for analysis of low-frequency and rare alleles, including patient-derived DNA variants.
• Experimental Design Optimization: Includes a power and sample size calculator that quantifies the tradeoff between sequencing coverage and number of biological replicates, noting that additional replicates reduce variance in effect size estimates via the Poisson-binomial distribution.

Scientific Applications:

• Regulatory variant identification in MPRAs: Detects and quantifies allelic regulatory effects in STARR-seq and other massively parallel reporter assays.
• Functional annotation of noncoding variants: Assigns functional significance to noncoding genetic variants by measuring their effects on reporter expression.
• Discovery of disease-causing mutations: Pinpoints causal regulatory mutations from patient-derived DNA, including low-frequency alleles.
• MPRA experimental planning: Informs sample size and sequencing coverage choices to optimize power for detecting allelic effects.

Methodology:

Implements a Bayesian hierarchical modeling framework that integrates prior information and models replicate variability; exploits independent locus assays to avoid linkage disequilibrium confounding; and performs power and sample size calculations using the Poisson-binomial distribution to evaluate tradeoffs between sequencing coverage and replicate number.