Predicting miRNA-disease associations based on lncRNA-miRNA interactions and graph convolution networks.

Area of Science:

• Biomedical informatics
• Computational biology
• Genomics

Background:

• MicroRNAs (miRNAs) are key biomarkers for human complex diseases.
• Predicting miRNA-disease associations (MDAs) aids disease prevention, diagnosis, and treatment.
• Existing computational methods for MDAs rely on similarity calculations, facing challenges due to data deficiencies.

Purpose of the Study:

• To develop a novel deep learning computational method, MAGCN, for predicting potential miRNA-disease associations (MDAs).
• To predict MDAs without relying on similarity measurements, addressing limitations of existing approaches.
• To identify novel disease-related miRNAs for improved disease management.

Main Methods:

• Proposed MAGCN, a deep learning model utilizing graph convolution networks with a multichannel attention mechanism and a convolutional neural network combiner.
• Leveraged known long non-coding RNA-miRNA interactions to predict MDAs.
• Conducted extensive experiments using 2-fold, 5-fold, and 10-fold cross-validations.

Main Results:

• Achieved high average area under the receiver operating characteristic (ROC) values: 0.8994 (2-fold), 0.9032 (5-fold), and 0.9044 (10-fold).
• Demonstrated superior prediction accuracy compared to five state-of-the-art methods.
• Case studies on three diseases confirmed that top predictions were supported by established databases.

Conclusions:

• MAGCN is a reliable computational tool for detecting novel disease-related miRNAs.
• The method effectively predicts miRNA-disease associations without using similarity measurements.
• Findings support the potential of MAGCN in advancing disease diagnosis and therapeutic strategies.