A multi-scale biological network framework for discovering MiRNA-Disease associations.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Identifying microRNA-disease associations is crucial for understanding complex diseases and finding biomarkers.
• Current computational methods struggle with sparse data, imbalanced disease distributions, and limited validated associations.

Purpose of the Study:

• To propose MMAG, a novel framework for microRNA-disease association prediction.
• To overcome limitations of existing methods using multi-scale biological graphs and meta-learning.

Main Methods:

• MMAG uses a multi-scale representation learning module to capture hierarchical biological information.
• A meta-learning strategy enables adaptation to few-shot settings by modeling diseases as individual tasks.
• Conditional adversarial alignment reduces feature distribution discrepancies for improved cross-task knowledge transfer.

Main Results:

• MMAG consistently outperforms state-of-the-art methods in few-shot, long-tailed, and cross-dataset scenarios.
• The framework demonstrates robust performance across various challenging prediction settings.

Conclusions:

• MMAG offers an effective and scalable solution for microRNA-disease association prediction.
• The proposed strategy shows promise for broader biological network inference tasks.