Identifying disease-related microbes based on multi-scale variational graph autoencoder embedding Wasserstein...

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Microbes play a critical role in human health.
• Identifying microbe-disease associations aids in understanding disease mechanisms, early diagnosis, and personalized medicine.
• Existing methods face challenges with data perturbation and suboptimal latent representations.

Purpose of the Study:

• To develop a novel framework for predicting disease-related microbes.
• To address limitations of data perturbation and improve latent representations.
• To enhance the prediction of microbe-disease associations for precision medicine.

Main Methods:

• Proposed a Multi-scale Variational Graph AutoEncoder embedding Wasserstein distance (MVGAEW) framework.
• Integrated multiple similarities using similarity network confusion.
• Employed an improved variational graph autoencoder for node latent representations and XGBoost for prediction.
• Introduced multi-order node embedding reconstruction to boost representation capacity.

Main Results:

• The MVGAEW framework effectively predicts disease-related microbes.
• The model demonstrates resilience to data perturbation and generates robust latent representations.
• Experiments on Alzheimer's disease, Crohn's disease, and colorectal neoplasms validated the framework's effectiveness.

Conclusions:

• The proposed MVGAEW model significantly outperforms existing state-of-the-art methods.
• Achieved substantial improvements on the HMDAD database.
• Highlights the potential of the framework for advancing microbe-disease association research.