BowSaw

BowSaw infers higher-order variable interactions from random forest models to classify phenotypes from genomic, transcriptomic, and metagenomic data.

Key Features:

• Random Forest Model Utilization: Uses a trained random forest and inspects decision tree structure to identify combinations of variables ("rules") frequently used for classification.
• Interpretation of Complex Data: Processes large-scale genomic, transcriptomic, and metagenomic datasets to extract higher-order interactions among variables.
• Enhanced Understanding of System Variables: Produces interpretable variable combinations that link specific system variables to classification outputs, enabling mechanistic insight.
• Application to Simulated and Real Datasets: Recovered variable sets used to generate phenotypes via complex Boolean rules in simulations despite noise, and identified high-order combinations of microbial taxa associated with Crohn's disease in the integrative Human Microbiome Project.

Scientific Applications:

• Phenotype Classification: Identifies key variable interactions for classifying cellular, organismal, and ecological phenotypes.
• Disease Association Studies: Detects high-order combinations of taxa or other variables associated with disease, as demonstrated for Crohn's disease.
• Hypothesis Generation: Extracts decision-tree-derived rules from random forests to generate testable biological hypotheses.

Methodology:

Analyzes the structure of decision trees within a trained random forest to identify frequently used variable combinations ("rules").