SFGAE: a self-feature-based graph autoencoder model for miRNA-disease associations prediction.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial biomarkers for diseases.
• Graph neural networks (GNNs) are used for miRNA-disease association prediction.
• Existing GNNs suffer from over-smoothing, hindering performance with deeper layers.

Purpose of the Study:

• To address the over-smoothing issue in GNN-based miRNA-disease association prediction.
• To propose a novel self-feature-based graph autoencoder model (SFGAE).
• To improve the accuracy and robustness of predicting miRNA-disease associations.

Main Methods:

• Developed SFGAE, a self-feature-based graph autoencoder.
• Constructed independent miRNA-self and disease-self embeddings.
• Employed a graph encoder with attention and a bilinear decoder for link prediction.

Main Results:

• SFGAE achieved state-of-the-art performance on benchmark datasets (HMDD v2.0, HMDD v3.2).
• SFGAE demonstrated superior performance with limited training data (e.g., 10%).
• SFGAE effectively resolved the over-smoothing problem, showing stable performance in deeper models.

Conclusions:

• SFGAE is a reliable tool for predicting potential miRNA-disease associations.
• The novel self-feature embeddings enhance prediction accuracy.
• SFGAE offers a robust solution for miRNA-disease association prediction, outperforming existing methods.