Graph neural network and multi-data heterogeneous networks for microbe-disease prediction

Area of Science:

• Microbiology
• Bioinformatics
• Computational Biology

Background:

• Understanding microbe-disease associations is crucial for pathogen mechanism research and precision medicine.
• Existing methods for predicting these associations require improvement in accuracy and efficiency.

Purpose of the Study:

• To develop and validate a novel computational model for predicting microbe-disease associations.
• To leverage multi-data biological networks and graph neural networks for enhanced prediction accuracy.

Main Methods:

• Constructed a Microbe-Disease Heterogeneous Network using microbe similarity, disease similarity, and known associations from the HMDAD database.
• Integrated this network into a graph convolutional neural network (GCNN) algorithm, developing the GCNN4Micro-Dis model.
• Evaluated model performance using 5-fold cross-validation and compared it against established methods (KATZHMDA, BiRWHMDA, LRLSHMDA).

Main Results:

• The GCNN4Micro-Dis model achieved an average AUC of 0.8954 ± 0.0030, indicating strong predictive power.
• The model outperformed three advanced methods in predicting microbe-disease associations.
• A case study on breast cancer identified 12 relevant microbes from patient gut flora, validating the model's practical accuracy.

Conclusions:

• The GCNN4Micro-Dis model offers a robust and accurate approach for predicting microbe-disease associations.
• This tool can significantly aid in discovering novel microbial links to diseases.
• The findings support the application of advanced computational models in precision medicine and microbial research.