BitSeq

BitSeq estimates transcript-level expression and differential expression from RNA-seq data using a two-stage Bayesian methodology.

Key Features:

• Two-stage Bayesian framework: Separates transcript-level expression estimation and differential expression inference into two Bayesian stages.
• Transcript-level Bayesian inference: Uses Bayesian inference to estimate expression of individual transcripts while accounting for ambiguities from shared exons and finite read sampling.
• MCMC posterior sampling: Represents relative expression via Markov chain Monte Carlo (MCMC) samples drawn from the posterior distribution of a generative sequencing model.
• Generative sequencing model: Employs a generative model tailored to sequencing data to model read generation and uncertainty.
• Replicate-aware DE analysis: Performs differential expression analysis across multiple conditions by leveraging replicates and propagating uncertainty from sample-level models.
• Expression-level-dependent prior: Incorporates biological variance through an expression-level-dependent prior to improve robustness of differential expression estimates.
• Handling of technical and biological variation: Models both technical and biological replication when estimating expression and differential expression.
• Validation on datasets: Demonstrated on both simulated data and real RNA-seq datasets.

Scientific Applications:

• Transcript expression quantification: Estimating expression levels of individual transcripts from RNA-seq experiments.
• Differential expression analysis: Identifying differential transcript expression across conditions using replicate data.
• Studies with technical and biological replication: Analyzing datasets that include technical and biological replicates to account for multiple sources of variance.
• Method benchmarking: Evaluating performance on simulated and real RNA-seq datasets.

Methodology:

Applies a two-stage Bayesian methodology: first stage uses Bayesian inference with MCMC sampling from a generative sequencing model to estimate transcript expression, second stage performs DE analysis across replicates by propagating sample-level uncertainty and applying an expression-level-dependent prior to model biological variance.