A comprehensive evaluation of multicategory classification methods for microbiomic data.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Next-generation DNA sequencing advances high-throughput microbial community analysis in human samples, establishing the field of microbiomics.
• Microbiomics holds potential for linking microbial taxa to health states, aiding diagnostics, personalized medicine, and forensics.
• Machine learning has shown promise in classifying microbiomic data, but optimal methods remain unclear.

Purpose of the Study:

• To systematically compare the performance of various machine learning classifiers for human microbiomic data analysis.
• To identify the most effective machine learning techniques for accurate classification tasks using microbial community composition.

Main Methods:

• Evaluated 18 major classification methods, 5 feature selection techniques, and 2 accuracy metrics.
• Utilized 8 distinct datasets comprising 1,802 human samples.
• Applied methods to classification tasks including body site, subject identification, and disease diagnosis.

Main Results:

• Identified specific machine learning algorithms as superior for microbiomic data classification.
• Demonstrated the effectiveness of certain methods across diverse classification challenges.
• Quantified classification performance using established accuracy metrics.

Conclusions:

• Random forests, support vector machines, kernel ridge regression, and Bayesian logistic regression with Laplace priors are highly effective for microbiomic classification.
• These methods offer robust performance for analyzing microbial community composition in human samples.
• The findings guide the selection of machine learning tools for advancing microbiomic research and applications.