Phylogeny-informed random forests for human microbiome studies.

Area of Science:

• Microbiome Research
• Bioinformatics
• Machine Learning

Background:

• Human microbiome studies generate complex, skewed, and irregular data.
• Traditional Random Forest algorithms may not fully leverage microbial feature relationships.
• Phylogenetic information offers insights into microbial evolutionary ancestry and functional relatedness.

Purpose of the Study:

• To introduce Phylogeny-Informed Random Forests (PIRF), an extension of Random Forest.
• To improve predictive accuracy in human microbiome studies by incorporating phylogenetic tree information.
• To provide a valuable tool for microbiome-based disease diagnostics and personalized medicine.

Main Methods:

• Developed PIRF, a Random Forest extension integrating phylogenetic tree information.
• PIRF employs a localized approach, identifying informative features within phylogenetic clusters.
• Evaluated PIRF's performance on seven benchmark tasks, including classification and regression problems.

Main Results:

• PIRF demonstrated high predictive accuracy across diverse human microbiome datasets.
• The localized approach of PIRF enriches functional representations and reduces tree-to-tree correlation.
• PIRF outperformed other standard tools in benchmark comparisons.

Conclusions:

• PIRF offers improved predictive performance for human microbiome analysis.
• Integrating phylogenetic information enhances the utility of Random Forest for microbiome data.
• PIRF is available as an R package for broader application in research and clinical settings.