MGCNSS: miRNA-disease association prediction with multi-layer graph convolution and distance-based negative sample...

Area of Science:

• Biomedical informatics
• Computational biology
• Genomics

Background:

• MicroRNAs (miRNAs) play crucial roles in disease mechanisms, making their identification vital for developing new therapeutics.
• Network-based methods are common for inferring miRNA-disease associations, but often overlook meta-path importance and reliable negative sample selection.

Purpose of the Study:

• To propose MGCNSS, a novel approach utilizing multi-layer graph convolution and a high-quality negative sample selection strategy.
• To improve the accuracy of predicting potential associations between microRNAs and diseases.

Main Methods:

• Constructing a comprehensive heterogeneous network integrating miRNA and disease similarity with known associations.
• Employing multi-layer graph convolution to capture diverse meta-path relations and learn miRNA/disease representations.
• Implementing a negative distance-based sample selection strategy to create a reliable negative sample set.

Main Results:

• MGCNSS demonstrated superior performance compared to baseline methods on both balanced and imbalanced datasets.
• Case studies on colon and esophageal neoplasms validated MGCNSS's capability in identifying potential candidate miRNAs.

Conclusions:

• MGCNSS offers an effective approach for predicting miRNA-disease associations by leveraging multi-layer graph convolutions and robust negative sampling.
• The method holds promise for advancing our understanding of disease mechanisms and facilitating the discovery of novel therapeutic targets.