Predicting miRNA-disease associations based on graph attention network with multi-source information.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) regulate cellular activities and disease processes.
• Identifying miRNA-disease associations is crucial for understanding pathogenesis and treatment.
• Current computational methods struggle with data sparsity and feature limitations.

Purpose of the Study:

• To develop an advanced computational framework for discovering novel miRNA-disease associations.
• To overcome limitations of existing models in predicting miRNA-disease relationships.
• To improve the accuracy and utility of computational approaches in this field.

Main Methods:

• A novel framework, GATMDA, was developed.
• It integrates multi-source information, fusing linear and non-linear features.
• Graph attention networks and random forest algorithms were employed for feature extraction and inference.

Main Results:

• GATMDA achieved a high average AUC of 0.9566 in five-fold cross-validation.
• Performance surpassed previous computational models.
• Case studies validated top predictions for breast cancer, colon cancer, and lymphoma with high accuracy.

Conclusions:

• GATMDA demonstrates significant accuracy and utility in identifying unknown miRNA-disease associations.
• The framework shows promise as a valuable tool for researchers.
• Results support the potential of GATMDA for advancing disease mechanism research.