Hierarchical Feature Engineering

Hierarchical Feature Engineering leverages phylogenetic hierarchies to engineer generalized features from microbiome metagenomic data for improved classification accuracy and biological interpretation.

Key Features:

• Phylogenetic Hierarchy Utilization: Exploits phylogenetic hierarchy/taxonomy to engineer features that capture traits distributed mono- or oligophyletically across microbial taxa.
• Reduction in Feature Space: Generates generalized, taxonomy-derived features that reduce dimensionality of the feature space while preserving biologically meaningful signals.
• Improved Classification Accuracy: Enhances classification performance within microbiota classification pipelines across diverse datasets independent of specific machine learning models.
• Explanatory Value of Features: Produces concise, interpretable features that provide pathophysiological insights into microbial community conditions aligned with expert analyses.

Scientific Applications:

• Metagenomic case-control classification: Distinguishes healthy versus diseased microbiota samples using taxonomy-informed features derived from metagenomic data.
• Disease-focused microbiome research: Supports studies of diseases with microbial components, including inflammatory bowel disease, obesity, and diabetes, by improving accuracy and interpretability of microbiome analyses.

Methodology:

The algorithm is embedded in a microbiota classification pipeline that computes relative abundances of operational taxonomic units (OTUs) from phylogenetic marker gene profiles and then applies hierarchical feature engineering to abstract and refine features according to phylogenetic relationships.