Using convolutional neural networks to explore the microbiome.

Area of Science:

• Microbiology
• Genomics
• Computational Biology

Background:

• The host microbiome significantly influences disease development.
• Accurate prediction of host phenotype from microbial data is crucial for personalized medicine.

Purpose of the Study:

• To evaluate a deep learning framework, specifically a convolutional neural network (CNN), for predicting host phenotype based on microbial taxonomic abundance profiles.
• To explore phylogenetic tree structures and develop an embedding scheme to enhance CNN learning from abundance data.

Main Methods:

• Utilized a convolutional neural network (CNN) deep learning framework.
• Constructed phylogenetic trees from microbial abundance profiles.
• Designed a matrix embedding scheme to preserve tree structure and quantitative characteristics for CNN input.

Main Results:

• Achieved a high prediction accuracy of 99.47% for host phenotype classification.
• Demonstrated the effectiveness of the proposed CNN framework on a dataset of 1967 samples across three phenotypes.

Conclusions:

• The study validates the feasibility and effectiveness of using CNNs for classifying sample phenotypes based on microbiome data.
• The findings suggest a promising approach for advancing personalized medicine through microbiome analysis.