Predicting miRNA-Disease Association Based on Neural Inductive Matrix Completion with Graph Autoencoders and...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are implicated in numerous human diseases.
• Accurate prediction of miRNA-disease associations is crucial for understanding disease mechanisms and developing treatments.
• Existing computational methods face challenges in effectively integrating matrix completion and label propagation.

Purpose of the Study:

• To propose a novel computational method, NIMGSA, for predicting potential miRNA-disease associations.
• To enhance the accuracy and robustness of miRNA-disease association prediction by unifying inductive matrix completion and label propagation.
• To leverage graph autoencoders (GAE) and self-attention mechanisms within a neural network framework.

Main Methods:

• Developed Neural Inductive Matrix completion with Graph Autoencoders (GAE) and Self-Attention (NIMGSA).
• Employed collaborative training of two GAEs to unify matrix completion and label propagation.
• Implemented a self-attention mechanism within the neural network architecture.
• Utilized cross-validation and case studies for performance evaluation.

Main Results:

• NIMGSA demonstrated superior performance compared to existing miRNA-disease prediction methods in cross-validation tests.
• Case studies confirmed the method's capability in identifying novel and potential miRNA-disease associations.
• The integrated framework enhanced the robustness and precision of both matrix completion and label propagation.

Conclusions:

• NIMGSA offers a robust and precise end-to-end framework for predicting miRNA-disease associations.
• The method effectively combines inductive matrix completion and label propagation, outperforming previous approaches.
• NIMGSA holds significant potential for advancing disease pathogenesis research and therapeutic strategies.