KGNMDA: A Knowledge Graph Neural Network Method for Predicting Microbe-Disease Associations.

Area of Science:

• Microbiology
• Computational Biology
• Bioinformatics

Background:

• Growing evidence links human microbes to complex diseases, highlighting their importance in drug development.
• Existing computational methods for predicting microbe-disease associations often lack comprehensive biological information.
• There is a need for advanced models that integrate diverse data for accurate association prediction.

Purpose of the Study:

• To develop an advanced computational model for predicting microbe-disease associations.
• To leverage a knowledge graph and graph neural networks for enhanced prediction accuracy.
• To provide a robust tool for understanding the intricate relationships between microorganisms and human diseases.

Main Methods:

• Constructed a knowledge graph integrating microbial and disease data from multiple databases.
• Employed a graph neural network (GNN) to learn representations of microbes and diseases.
• Incorporated Gaussian kernel similarity features to refine representations and predict associations.
• Utilized a score function for quantifying microbe-disease association probabilities.

Main Results:

• The proposed model significantly outperformed existing baseline methods on the Human Microbe-Disease Association Database (HMDAD).
• Case studies on asthma and inflammatory bowel disease demonstrated the model's effectiveness in revealing disease-microbe relationships.
• The model successfully identified potential microbe-disease associations, validating its predictive power.

Conclusions:

• The developed knowledge graph-based neural network model provides an effective approach for predicting microbe-disease associations.
• This method enhances the understanding of the role of microbes in human diseases.
• The findings offer valuable insights for targeted drug development and personalized medicine strategies.