A Comprehensive Study on Predicting Functional Role of Metagenomes Using Machine Learning Methods.

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Metagenomics involves studying microbial communities from environmental samples.
• High-throughput sequencing generates vast, complex datasets, leading to the 'curse of dimensionality'.
• Machine learning (ML) is crucial for analyzing high-dimensional metagenomic data efficiently.

Purpose of the Study:

• To assess ML approaches for classifying human microbiota into functional phenotypes.
• To introduce embedded feature selection methods for identifying key microbial species.
• To enhance the performance of existing ML models in metagenomic studies.

Main Methods:

• Applied Extreme Gradient Boosting and Penalized Logistic Regression for feature selection.
• Utilized the selected features to improve the Random Forest (RF) classifier.
• Compared the proposed method against other feature selection techniques and classifiers (SVM, ELM, k-NN).

Main Results:

• The proposed embedded feature selection method significantly improved RF classifier performance.
• Achieved superior accuracy and Receiver Operating Characteristic Area Under the Curve (ROC-AUC).
• Demonstrated a substantial reduction in data processing time compared to other methods.

Conclusions:

• Embedded feature selection methods are effective for high-dimensional human microbiota classification.
• The proposed approach enhances accuracy and efficiency in metagenomic data analysis.
• This method offers a computationally efficient solution for linking microbial structures to functional roles.