edgeR

edgeR performs differential expression analysis of RNA-seq and other genomic count-based data (e.g., ChIP-seq, SAGE, CAGE) by using negative binomial models to estimate gene-wise variability and test for expression changes across complex experimental designs.

Key Features:

• Statistical Framework: Implements negative binomial models that accommodate complex experimental designs with multiple treatment conditions and blocking variables to model biological and technical variation.
• Empirical Bayes Methods: Uses empirical Bayes approaches to estimate and moderate gene-specific dispersions, improving variability estimates with limited biological replicates.
• Generalized Linear Models (GLMs): Supports GLM-based testing with genewise dispersions for paired designs and identification of condition- or tumour-specific changes.
• Handling Overdispersion: Models overdispersion via an overdispersed Poisson/negative binomial approach with empirical Bayes moderation of dispersion estimates.
• Parallel Computational Approaches: Applies parallel computation to facilitate non-linear model fitting and to scale GLM applications to genomic datasets.
• Simulations and Estimators: Uses simulations to assess adjusted profile likelihood estimators and provides empirical Bayes estimators that balance common-dispersion and genewise-dispersion assumptions.

Scientific Applications:

• RNA-seq Analysis: Identifies differentially expressed genes and pathways from RNA-seq count data.
• ChIP-seq and Other Genomic Data: Applies count-based statistical methods to ChIP-seq, SAGE, CAGE and other genomic count data types, including analyses related to epigenetic marks and DNA methylation.
• Pathway Analysis: Facilitates downstream pathway-level analyses and visualization of differential expression results.

Methodology:

Integrates with R and Bioconductor, accepts counts from read alignment (e.g., Rsubread) and count quantification, and performs differential expression analysis using quasi-likelihood methods based on negative binomial models.