CBEA

CBEA performs set-based taxonomic enrichment analysis in R using the Competitive Balances for Taxonomic Enrichment Analysis (CBEA) competitive log-ratio formulation on compositional, sparse, high-dimensional microbiome relative abundance data to generate sample-specific enrichment scores.

Key Features:

• Novel Log-Ratio Formulation: Implements a competitive null hypothesis-based, scaled log-ratio of subcompositions defined by taxa within a set and its complement to compute sample-specific enrichment scores.
• Handling High-Dimensionality and Sparsity: Operates on taxonomic count tables from high-throughput sequencing of human-associated microbiomes and accommodates common high-dimensionality and sparsity in relative abundance data.
• Improved Aggregation Methodology: Avoids simple abundance summation for aggregation to enable comparison across set sizes, preserve inter-sample distance structure, and reduce amplification of protocol bias.
• Empirical Null Distribution: Estimates an empirical null distribution for the test statistic to provide sample-level p-values and control type I error under high sparsity and inter-taxa correlation.
• Informative Enrichment Scores: Produces enrichment scores suitable as inputs for downstream analyses, including predictive modeling in microbiome research.

Scientific Applications:

• Set-level enrichment detection: Identifies enriched taxonomic sets within complex microbiome datasets derived from high-throughput sequencing.
• Predictive modeling input: Supplies sample-specific enrichment scores as features for phenotype or disease prediction tasks in microbiome studies.
• Microbial community interpretation: Facilitates analysis of relative abundance and taxon interactions relevant to human health, disease mechanisms, and potential therapeutic interventions.

Methodology:

Computes scaled log-ratios of subcompositions for taxa in a set versus its complement under a competitive log-ratio formulation to generate single-sample enrichment scores and estimates an empirical null distribution for sample-level significance testing.