Predicting miRNA-disease associations based on PPMI and attention network.

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial in various diseases, but experimental identification of miRNA-disease associations (MDAs) is costly and slow.
• Efficient computational methods are needed to explore MDAs and understand complex disease pathogenesis.

Purpose of the Study:

• To develop an efficient computational method for predicting miRNA-disease associations (MDAs).
• To identify novel disease-related microRNAs (miRNAs) through computational analysis.

Main Methods:

• Constructed heterogeneous miRNA-disease association (MDA) and similarity networks.
• Employed random walk with restart and positive point-wise mutual information (PPMI) for feature extraction.
• Utilized convolutional neural networks and an attention network with neural aggregation for representation learning and integration.
• Applied an inner product decoder to predict relationship scores.

Main Results:

• The proposed PATMDA method achieved high predictive performance.
• Achieved an area under the receiver operating characteristic curve (AUC) of 0.933 on HMDD v2.0 and 0.946 on HMDD v3.2.
• Case studies validated PATMDA's ability to discover novel disease-associated miRNAs.

Conclusions:

• PATMDA demonstrates superior performance compared to existing state-of-the-art methods for predicting miRNA-disease associations.
• The method effectively aids in the discovery of novel disease-related miRNAs, offering a valuable tool for biomedical research.