dexus

dexus identifies differentially expressed transcripts in RNA-Seq datasets when sample conditions are unknown or not predefined.

Key Features:

• Modeling approach: Models read counts using finite mixtures of negative binomial distributions with each mixture component corresponding to an inferred condition.
• Bayesian framework and EM algorithm: Performs model selection and parameter estimation within a Bayesian framework using an Expectation-Maximization (EM) algorithm.
• Decomposition of variation: Separates variation attributable to noise from variation attributable to differential expression.
• Informative/Noninformative (I/NI) value: Computes an I/NI value to quantify evidence for differential expression and enable transcript selection at different specificity/sensitivity trade-offs.
• Performance metrics: In simulations of 2400 instances, reported specificity/sensitivity were: I/NI=0.025 → 92% specificity, 76% sensitivity; I/NI=0.05 → 97% specificity, 61% sensitivity; I/NI=0.1 → 99% specificity, 38% sensitivity.

Scientific Applications:

• Cohort and cross-sectional analyses: Analyzing RNA-Seq data from cohort studies and cross-sectional analyses with unknown sample conditions.
• Nonrandomized controlled trials: Analyzing RNA-Seq data from nonrandomized controlled trials without predefined sample conditions.
• Detection of biological factors: Detecting expression differences associated with sex, species, tissue types, structural variants, or quantitative trait loci (QTLs).

Methodology:

Finite mixtures of negative binomial distributions; Bayesian model selection and parameter estimation via an Expectation-Maximization (EM) algorithm; decomposition of expression variation into noise and differential expression; computation of the Informative/Noninformative (I/NI) value.