Synchronous Mutual Learning Network and Asynchronous Multi-Scale Embedding Network for miRNA-Disease Association...

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are key regulators of cellular processes, and identifying miRNA-disease associations (MDAs) is vital for disease comprehension.
• Graph Neural Networks (GNNs) show promise in MDA prediction, but often rely on single network representations, neglecting diverse data attributes.

Purpose of the Study:

• To propose SMDAP, a novel GNN-based framework for predicting MDAs by integrating multiple network topologies and node attributes.
• To leverage both miRNA seed and full-length sequences for enhanced MDA prediction accuracy.

Main Methods:

• SMDAP employs a dual-branch architecture: a synchronous mutual learning network using miRNA seed sequences and an asynchronous multi-scale embedding network using full-length sequences.
• An adaptive fusion approach combines representations from both branches to score potential MDAs.
• The framework utilizes GNNs to learn node representations from heterogeneous and homogeneous network patterns.

Main Results:

• SMDAP effectively integrates diverse network topologies and node attributes, outperforming existing methods.
• Achieved an Area Under the Receiver Operating Characteristic Curve (AUC) of 0.9622 and an Area Under the Precision-Recall Curve (AUPRC) of 0.9777 on DB1, showing significant improvements over baselines.
• Case studies on three human cancers validated the framework's predictive capabilities for novel MDAs.

Conclusions:

• SMDAP represents a significant advancement in MDA prediction by incorporating multi-topology and multi-attribute learning.
• The framework offers a robust and accurate tool for identifying potential miRNA-disease links, aiding in disease mechanism research.
• SMDAP's performance highlights the importance of diverse data integration in predictive biological modeling.