MUS-HGFC: Inferring miRNA-disease associations with multi-scale hypergraph representations.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are key regulators whose dysregulation is linked to diseases like cancer.
• Predicting miRNA-disease associations (MDAs) is vital for identifying disease biomarkers and therapeutic targets.
• Existing computational methods, like graph neural networks (GNNs), often oversimplify complex biological interactions.

Purpose of the Study:

• To develop a novel computational framework, MUS-HGFC, for accurate prediction of miRNA-disease associations.
• To address limitations of pairwise modeling in current GNN-based approaches.
• To leverage higher-order biological relationships for improved MDA prediction.

Main Methods:

• Developed a leakage-free miRNA similarity measure using an experimentally validated miRNA-target gene network.
• Modeled biological interactions using a hypergraph structure to represent higher-order relationships.
• Employed multi-scale hypergraph convolutional layers to learn node representations from integrated topological information.

Main Results:

• MUS-HGFC demonstrated superior performance compared to state-of-the-art methods in benchmarking experiments.
• The framework achieved a high validation rate for top-ranked predicted miRNA-disease associations in case studies.
• The approach effectively captures the complexity of biological networks for robust MDA prediction.

Conclusions:

• MUS-HGFC provides a powerful and robust framework for discovering novel miRNA-disease associations.
• The method's ability to model higher-order interactions enhances the accuracy of computational predictions.
• This work offers a valuable tool for advancing cancer research and personalized medicine through biomarker identification.