phylaGAN: data augmentation through conditional GANs and autoencoders for improving disease prediction accuracy using...

Area of Science:

• Microbiome research
• Computational biology
• Machine learning

Background:

• The human microbiome plays a crucial role in health and disease.
• Machine learning effectively distinguishes healthy from diseased microbiome states.
• Challenges in microbiome ML include small sample sizes, data imbalance, and high data collection costs.

Purpose of the Study:

• To propose a deep learning framework, phylaGAN, for augmenting microbiome datasets.
• To address limitations of small sample sizes and data imbalance in microbiome machine learning.
• To improve the accuracy of disease prediction models using microbiome data.

Main Methods:

• Developed phylaGAN, a deep learning framework combining conditional generative adversarial networks (C-GAN) and autoencoders.
• Utilized C-GAN to generate representative synthetic microbiome data, expanding existing datasets.
• Employed autoencoders to map original and generated samples into a common subspace for enhanced prediction.

Main Results:

• PhylaGAN demonstrated significant improvements in predictive accuracy across multiple datasets.
• Mean AUC increased by 11% in a T2D study and 5% in a Cirrhosis study.
• External validation on an obesity cohort showed a ~32% improvement in mean AUC with phylaGAN augmentation.

Conclusions:

• Generative adversarial networks can effectively create synthetic data that mimics real microbiome data.
• PhylaGAN shows potential for enhancing disease prediction by training machine learning models on augmented datasets.
• The framework offers a viable solution to data scarcity challenges in microbiome research.