Variational graph auto-encoders for miRNA-disease association prediction.

Area of Science:

• Genomics
• Computational Biology
• Biomedical Informatics

Background:

• MicroRNAs (miRNAs) play crucial roles in biological processes and human diseases.
• Experimental identification of miRNA-disease associations is costly and time-consuming.
• Computational methods are needed for efficient prediction of these associations.

Purpose of the Study:

• To develop an effective deep learning framework for predicting miRNA-disease associations.
• To improve the accuracy and efficiency of identifying links between miRNAs and diseases.
• To provide a computational tool for understanding disease mechanisms and potential therapeutic targets.

Main Methods:

• A deep learning framework utilizing a variational graph auto-encoder (VGAE-MDA) was developed.
• Heterogeneous networks integrating miRNA-miRNA similarity, disease-disease similarity, and known associations were constructed.
• Two sub-networks (miRNA-based and disease-based) were created, each employing a VGAE for score calculation.
• Graph convolutional networks (GCN) and variational autoencoders (VAE) were used to incorporate node features and predict associations from data distribution.

Main Results:

• VGAE-MDA demonstrated superior performance compared to state-of-the-art methods in miRNA-disease association prediction.
• The model effectively mitigates noise from negative sample selection.
• Case studies validated the practical effectiveness of the VGAE-MDA predictions.

Conclusions:

• VGAE-MDA offers a powerful computational approach for predicting miRNA-disease associations.
• The framework enhances understanding of disease mechanisms and aids in diagnosis and treatment strategies.
• This deep learning model represents a significant advancement in the field of computational biology for disease association studies.