Predicting miRNA-disease associations using improved random walk with restart and integrating multiple similarities

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Predicting miRNA-disease associations (MDAs) is crucial but experimentally challenging.
• Computational methods are essential for efficient MDA prediction.
• Existing methods face challenges with data sparsity and incompleteness.

Purpose of the Study:

• To propose a novel computational method, RWRMMDA, for predicting miRNA-disease associations.
• To address data sparsity using a WKNKN algorithm as a pre-processing step.
• To enhance prediction accuracy by integrating multiple similarity measures within heterogeneous networks.

Main Methods:

• Developed a Weighted K-Nearest Neighbors (WKNKN) algorithm to handle data sparsity.
• Constructed heterogeneous networks for miRNA and disease spaces by integrating multiple similarity networks.
• Applied an improved random walk with restart (RWR) algorithm on these networks to calculate MDA probabilities.

Main Results:

• Achieved high performance with Global LOOCV AUC of 0.9882 and AUPR of 0.9066.
• Fivefold cross-validation yielded AUC of 0.9855 and AUPR of 0.8642, statistically significant.
• Outperformed existing methods (NTSHMDA, PMFMDA, IMCMDA, MCLPMDA) in AUC and AUPR.
• Case studies identified novel, literature-confirmed associations for Breast Neoplasms, Hepatocellular Carcinoma, and Stomach Neoplasms.

Conclusions:

• RWRMMDA is an effective computational method for predicting miRNA-disease associations.
• The integration of multiple similarities and improved RWR significantly enhances prediction accuracy.
• The method successfully identifies novel MDAs, validated by existing databases and literature.