Phylogenetic approaches to microbial community classification.

Area of Science:

• Microbiology
• Bioinformatics
• Machine Learning

Background:

• Microbiota composition varies across body sites, with subtle differences within sites like the mouth.
• Accurate predictive models are crucial for distinguishing oral sub-sites and understanding microbial roles.
• Machine learning (ML) offers potential for supervised classification, with feature representation being critical for model performance.

Purpose of the Study:

• To develop and evaluate ML models for classifying nine oral sites and dental plaque.
• To investigate the utility of phylogenetic information and predicted functions in feature representation for oral microbiota classification.
• To assess the impact of different feature representations and classifiers on prediction accuracy.

Main Methods:

• Utilized data from the Human Microbiome Project for oral microbiota analysis.
• Employed phylogenetic information for feature representation, including custom kernels and PICRUSt software for functional predictions.
• Compared various feature representations (clades, functions) and ML classifiers to predict oral site classification.

Main Results:

• Feature representation significantly impacted classification accuracy; microbial clade and function features were informative.
• Combining clade and function features did not improve prediction accuracy over using them individually.
• The best classification accuracy achieved for the dental plaque dataset was approximately 81%.

Conclusions:

• Classifying oral microbiota sub-sites remains challenging, with perfect accuracy potentially limited by site proximity and individual variation.
• Feature representation is a critical factor for improving predictive model accuracy in oral microbiome research.
• Further exploration of ML classifiers and feature representation strategies is recommended to enhance prediction accuracy.