Combined embedding model for MiRNA-disease association prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) play crucial roles in diagnosing and treating complex diseases.
• Traditional experimental methods for identifying miRNA-disease interactions are time-consuming and costly.
• Computational approaches are valuable for discovering potential miRNA-disease associations to aid disease pathogenesis and therapy.

Purpose of the Study:

• To develop a novel computational model, CEMDA, for predicting miRNA-disease associations.
• To enhance the accuracy of miRNA-disease interaction prediction by integrating combined embeddings.
• To improve upon existing heterogeneous network methods for miRNA-disease association discovery.

Main Methods:

• Constructed a heterogeneous network incorporating miRNA-disease pairs, disease semantic similarity, and miRNA functional similarity.
• Utilized Gate Recurrent Unit (GRU) with a multi-head attention mechanism to learn similarity measures from meta-paths.
• Integrated pair embedding, processed by a Multi-Layer Perceptron (MLP), with node embedding for comprehensive association prediction.

Main Results:

• The proposed Combined Embedding Model for MiRNA-disease Associations (CEMDA) effectively predicts miRNA-disease interactions.
• CEMDA achieved high accuracy rates of 93.16% (leave-one-out) and 92.03% (fivefold cross-validation).
• Validation on lung, breast, prostate, and pancreatic cancers demonstrated the model's effectiveness in identifying top miRNA candidates.

Conclusions:

• CEMDA demonstrates superior performance compared to existing methods for predicting miRNA-disease associations.
• The model's predictions are feasible and effective, as evidenced by cross-validation results and case studies.
• CEMDA offers a promising computational tool for advancing miRNA-based disease diagnosis and therapy.