BANDITS

BANDITS detects differential transcript usage and alternative splicing between conditions in RNA-seq data at gene and transcript levels using a Bayesian hierarchical model that accounts for sample-to-sample variability and mapping uncertainty.

Key Features:

• Implementation: Provided as an R/Bioconductor package for integration with Bioconductor workflows.
• Bayesian hierarchical framework: Models sample-specific proportions with a Dirichlet-Multinomial component to capture biological variability across samples.
• Latent variable modeling: Treats transcript allocation of reads as latent variables to account for mapping uncertainty of RNA fragments to transcripts.
• Markov Chain Monte Carlo (MCMC) inference: Uses MCMC techniques for parameter estimation and sampling from complex posterior distributions.
• Multivariate Wald test: Applies a multivariate Wald test on posterior densities of average relative transcript abundances to assess differential transcript usage.

Scientific Applications:

• Alternative splicing analysis: Identification of differential splicing events and differential transcript usage (DTU) between experimental conditions using RNA-seq data.
• Gene expression regulation studies: Investigation of transcript-level regulatory changes that affect gene expression profiles across conditions.
• Disease and developmental biology research: Detection of splicing alterations relevant to disease mechanisms and developmental processes.

Methodology:

BANDITS employs a Bayesian hierarchical model with a Dirichlet-Multinomial component for sample-specific proportions, treats read-to-transcript allocation as latent variables, performs parameter inference via MCMC, and assesses differential transcript usage using a multivariate Wald test on posterior densities of average relative transcript abundances.