ALDEx2

ALDEx2 performs differential abundance analysis of high-throughput sequencing count data by modeling compositionality with a Dirichlet-multinomial framework.

Key Features:

• Dirichlet-multinomial model: Transforms raw counts into relative abundances to handle compositional data and account for technical and statistical variation.
• Statistical tests: Implements Wilcox rank test, Welch's t-test, generalized linear models (GLM), and Kruskal-Wallis test for inference of differential abundance.
• False Discovery Rate (FDR) control: Calculates expected false discovery rate and reports P-values and FDR values adjusted using the Benjamini-Hochberg correction, considering biological and sampling variation.
• Bayesian inference: Uses Bayesian methods to distinguish technical noise from true biological signal.
• Replicate optimization: Model optimized for experiments with three or more replicates.

Scientific Applications:

• RNA Sequencing (RNA-seq): Identifies differentially expressed genes from count-based RNA-seq data.
• 16S rRNA Gene Sequencing: Analyzes microbial community taxon abundances, including distinguishing taxa between tongue dorsum and buccal mucosa in human microbiome studies.
• Chromatin Immunoprecipitation Sequencing (ChIP-seq): Applicable to epigenetic studies of DNA–protein interactions using count data.
• Metagenomic Analysis: Facilitates differential abundance analysis of microbial communities from environmental or clinical metagenomic samples.
• Selective Growth Experiments: Assesses differential growth patterns in vitro using count-based measurements.
• Human Microbiome Project 16S data: Has been applied to Human Microbiome Project 16S rRNA gene abundance datasets.

Methodology:

Applies a Dirichlet-multinomial framework to transform raw counts into relative abundances, uses Bayesian inference to partition technical noise and biological signal, performs statistical tests (Wilcox rank test, Welch's t-test, GLM, Kruskal-Wallis), and computes P-values and FDR values with Benjamini-Hochberg adjustment while estimating expected false discovery rate accounting for biological and sampling variation; optimized for experiments with three or more replicates.