Predicting miRNA-disease associations via layer attention graph convolutional network model.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are key regulators of gene expression, making miRNA-disease association prediction crucial for understanding disease mechanisms.
• Existing methods for predicting associations between miRNAs and diseases require improvement in accuracy and scope.
• The development of computational methods is essential for identifying potential miRNA-disease links.

Purpose of the Study:

• To introduce and evaluate the LAGCN method for identifying potential miRNA-disease associations.
• To leverage graph convolution networks and attention mechanisms for enhanced prediction accuracy.
• To provide a robust computational tool for miRNA-disease association discovery.

Main Methods:

• Constructed a heterogeneous network integrating known miRNA-disease associations, miRNA-miRNA similarities, and disease-disease similarities.
• Employed a graph convolution network (GCN) to learn low-dimensional embeddings for miRNAs and diseases.
• Utilized an attention mechanism to combine embeddings from multiple GCN layers for improved representation.

Main Results:

• The LAGCN method achieved a high Area Under the Curve (AUC) value of 0.9091 after fivefold cross-validation.
• LAGCN demonstrated superior performance compared to other established prediction methods and baseline approaches.
• The integrated embedding approach effectively captured complex relationships within the heterogeneous network.

Conclusions:

• LAGCN is an effective and accurate method for predicting potential miRNA-disease associations.
• The proposed method outperforms existing approaches, offering a valuable tool for biomedical research.
• This work highlights the potential of graph-based deep learning models in uncovering novel biological relationships.