Multi-Kernel Graph Attention Deep Autoencoder for MiRNA-Disease Association Prediction.

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• MicroRNAs (miRNAs) regulate biological processes, and their abnormal expression is linked to complex diseases.
• Identifying miRNA-disease associations (MDAs) is crucial for disease diagnosis and treatment.
• Experimental verification of MDAs is time-consuming and limited in scale, necessitating computational approaches.

Purpose of the Study:

• To propose a reliable and effective computational method for predicting novel miRNA-disease associations (MDAs).
• To develop a multi-kernel graph attention deep autoencoder (MGADAE) for enhanced MDA prediction accuracy.

Main Methods:

• Utilized multiple kernel learning (MKL) to integrate miRNA and disease similarity.
• Constructed a heterogeneous network incorporating known MDAs, disease similarity, and miRNA similarity.
• Employed graph convolution operations and an attention mechanism within a deep autoencoder framework (MGADAE) for feature learning and representation integration.
• Applied a bilinear decoder to predict final association scores.

Main Results:

• The proposed MGADAE method demonstrated superior performance compared to existing approaches in predicting MDAs.
• Experimental results validated the method's effectiveness and reliability.
• Case studies on human cancers further confirmed the practical utility of MGADAE.

Conclusions:

• MGADAE offers a powerful computational tool for predicting potential miRNA-disease associations.
• The method facilitates the discovery of novel MDAs, contributing to advancements in human disease diagnosis and treatment.
• This approach addresses the limitations of traditional experimental methods for MDA identification.