DeepMicro: deep representation learning for disease prediction based on microbiome data.

Area of Science:

• Microbiome research
• Computational biology
• Machine learning applications

Background:

• Human microbiota is crucial for health, with potential as a disease predictor.
• High-dimensional microbiome data and low sample sizes challenge machine learning models.
• Sparse data hinders effective prediction model development.

Purpose of the Study:

• To introduce DeepMicro, a deep representation learning framework for microbiome data.
• To address challenges of high dimensionality and sparsity in microbiome datasets.
• To improve disease prediction accuracy using microbiome profiles.

Main Methods:

• Deep representation learning using autoencoders to create low-dimensional microbiome profiles.
• Application of machine learning classification on learned representations.
• Rigorous evaluation across multiple datasets.

Main Results:

• DeepMicro outperforms existing methods in disease prediction across five datasets.
• Significant dimensionality reduction of microbiome data.
• Achieved 8X-30X speedup in model training and hyperparameter optimization.

Conclusions:

• DeepMicro offers an effective solution for analyzing high-dimensional microbiome data.
• The framework enhances disease prediction accuracy and computational efficiency.
• DeepMicro provides a valuable tool for microbiome-based health research.