MADGAN:A microbe-disease association prediction model based on generative adversarial networks

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Microorganisms are vital for human health, and their imbalance is linked to diseases.
• Understanding microbe-disease associations is crucial for medical research and treatment.
• Existing methods for predicting these associations have limitations.

Purpose of the Study:

• To propose a novel prediction model, MADGAN, for inferring potential microbe-disease associations.
• To leverage biological information of microbes and diseases with generative adversarial networks (GANs).
• To be the first to apply GANs for predicting microbe-disease associations.

Main Methods:

• Constructing microbe and disease features using multiple similarity metrics.
• Employing graph convolutional neural networks (GCN) for automatic feature derivation.
• Training MADGAN using a generative adversarial approach with a cross-level weight distribution structure to prevent over-smoothing.

Main Results:

• MADGAN successfully inferred latent microbe-disease associations.
• The model achieved satisfactory prediction performance in comprehensive experiments.
• MADGAN outperformed existing state-of-the-art prediction models on HMDAD and Disbiome databases.

Conclusions:

• MADGAN is an effective and novel model for predicting microbe-disease associations.
• The integration of GANs and GCN offers a powerful approach for this task.
• The findings highlight the potential of computational methods in understanding host-microbe interactions and disease.