BNBR

BNBR implements Bayesian negative binomial regression to model RNA-seq count data and infer differential expression in complex multi-factor experimental designs while accounting for overdispersion and confounding variables.

Key Features:

• Bayesian Negative Binomial Regression (BNB-R): Models RNA-seq count data using a Bayesian negative binomial framework that accounts for overdispersion.
• Handling Multi-Factor Experimental Designs: Accommodates complex multi-factor conditions and multivariate dependence structures without requiring normalization preprocessing.
• Efficient Bayesian Inference: Employs novel data augmentation techniques and exploits conditional conjugacy to achieve computationally efficient Bayesian parameter estimation.
• Natural Model Parameterization: Uses a parameterization that obviates the need for separate normalization steps across experimental setups.
• Performance Validation: Validated on synthetic and real RNA-seq datasets with superior areas under receiver operating characteristic and precision-recall curves.
• Implementation: Implemented in the R programming language.

Scientific Applications:

• Differential Expression Analysis: Identifies differentially expressed genes across multiple conditions while accounting for confounding factors.
• Genotype-Phenotype Relationship Deciphering: Supports complex experimental designs aimed at elucidating genotype-phenotype relationships.

Methodology:

Bayesian negative binomial regression with a natural parameterization, novel data augmentation techniques exploiting conditional conjugacy for efficient Bayesian inference, and performance assessment using AUROC and precision-recall curves on synthetic and real RNA-seq data.