Deep ensemble learning over the microbial phylogenetic tree (DeepEn-Phy).

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• The human microbiome is a key biomarker for predicting host clinical outcomes.
• Incorporating microbial phylogeny, the evolutionary relationships among microbes, can enhance prediction accuracy.
• Current methods may not fully exploit the rich information contained within microbial evolutionary history.

Purpose of the Study:

• To propose a novel deep neural network (PhyNN) and an ensemble method (DeepEn-Phy) for host clinical outcome prediction.
• To develop a method that optimally extracts features from microbial phylogeny, going beyond taxonomic information.
• To demonstrate the superiority of DeepEn-Phy in predicting both categorical and continuous clinical outcomes using large-scale microbiome data.

Main Methods:

• Development of a phylogeny-driven deep neural network (PhyNN).
• Creation of an ensemble method, DeepEn-Phy, integrating multiple PhyNN models.
• Application of DeepEn-Phy to a large real-world microbiome dataset for outcome prediction.

Main Results:

• DeepEn-Phy achieved superior prediction performance compared to existing machine learning and deep learning approaches.
• The method effectively extracted and utilized phylogenetic features for improved predictive power.
• Successful prediction of both categorical and continuous clinical outcomes was demonstrated.

Conclusions:

• DeepEn-Phy offers a new, powerful strategy for analyzing phylogeny-constrained microbiome data.
• Leveraging microbial phylogeny within deep learning architectures significantly enhances clinical outcome prediction.
• This work provides a framework for future deep neural network designs in microbiome research.