Predicting Potential Microbe-Disease Associations Based on Heterogeneous Graph Random Attention Neural Network and...

Area of Science:

• Microbiology
• Computational Biology
• Genomics

Background:

• Microbial communities play a crucial role in human diseases, but traditional identification methods are costly and time-consuming.
• Understanding microbe-disease associations (MDAs) is vital for disease mechanism elucidation and therapeutic development.

Purpose of the Study:

• To develop an efficient deep learning framework, GRNCFMDA, for predicting microbe-disease associations (MDAs).
• To overcome the limitations of traditional methods in terms of cost, time, and manual effort.

Main Methods:

• Constructed a heterogeneous network by integrating microbe and disease similarities (functional, GIP, semantic).
• Employed a graph random neural network (GRAND) with attention for node representation learning.
• Utilized a neural collaborative filtering module combining matrix factorization and multilayer perceptrons.

Main Results:

• GRNCFMDA demonstrated superior performance compared to four existing MDA prediction models across HMDAD and Disbiome datasets.
• Five-fold cross-validation confirmed the model's predictive accuracy.
• Case studies validated GRNCFMDA's effectiveness in identifying novel MDAs.

Conclusions:

• GRNCFMDA offers an efficient and accurate approach for predicting microbe-disease associations.
• The framework has practical utility in advancing research on the human microbiome and disease.
• Publicly available implementation and datasets facilitate further research and application.