DGAMDA: Predicting miRNA-disease association based on dynamic graph attention network.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Traditional experimental validation for microRNA (miRNA)-disease associations is time-consuming and costly.
• Computational methods leveraging artificial intelligence offer efficient alternatives for predicting miRNA-disease associations.
• Existing methods often struggle with feature heterogeneity and insufficient feature mining from static networks.

Purpose of the Study:

• To propose a dynamic graph attention-based association prediction model (DGAMDA) for enhanced miRNA-disease association prediction.
• To address limitations of static graph attention mechanisms in feature mining and heterogeneity.
• To achieve high-precision feature mining and association scoring using a single miRNA-disease association network.

Main Methods:

• Developed DGAMDA, a model integrating feature mapping and dynamic graph attention mechanisms.
• Employed feature mining on a single miRNA-disease association network.
• Conducted rigorous five-fold cross-validation experiments to evaluate model performance.

Main Results:

• DGAMDA demonstrated high-precision feature mining and association scoring.
• Achieved mean Accuracy of 0.8986, Precision of 0.8869, Recall of 0.9115, and F1-score of 0.8984.
• Outperformed other advanced models in prediction accuracy and effectiveness.

Conclusions:

• DGAMDA effectively overcomes feature heterogeneity and inadequate mining issues present in static models.
• The model shows significant promise for accurate miRNA-disease association prediction and early disease screening.
• DGAMDA can be utilized for predicting miRNAs associated with novel or unknown diseases.