Trends in Machine Learning and Feature Selection Stability for Human Gut Microbiome (Shotgun Metagenomics) and Metabolomics Matched Datasets

Machine learning models, especially Random Forest and XGBoost, achieve up to 20% higher feature selection stability in human gut microbiome multi-omics data.

• Why it matters: Reproducibility and functional insights in microbiome research are limited by inconsistent methods and unstable biomarker identification, hindering progress in understanding microbial functions.
• What they did: The study systematically benchmarked three algorithms across seven multi-omics strategies using human gut datasets, evaluating impacts of data transformation and feature reduction on prediction and stability.
• The result: Findings reveal that algorithm choice, integration approach, and data preprocessing significantly influence model accuracy and feature stability, enabling more reliable microbiome biomarker discovery.