Novel Human miRNA-Disease Association Inference Based on Random Forest.

Area of Science:

• Computational biology
• Genomics
• Biomedical informatics

Background:

• MicroRNAs (miRNAs) are increasingly linked to human complex diseases.
• Understanding miRNA-disease associations is crucial for advancing disease treatment.
• Traditional experimental methods for identifying these associations are costly and time-consuming.

Purpose of the Study:

• To develop a reliable computational model for predicting miRNA-disease associations.
• To leverage machine learning for efficient identification of potential miRNA-disease links.
• To improve upon existing computational methods for miRNA-disease association prediction.

Main Methods:

• Developed a Random Forest model for miRNA-disease association (RFMDA) prediction.
• Constructed training data using the Human microRNA Disease Database (HMDD) v2.0.
• Defined feature vectors by integrating miRNA functional similarity, disease semantic similarity, and Gaussian interaction profile kernel similarity.
• Employed a filter-based method for robust feature selection.

Main Results:

• RFMDA achieved high AUC values in cross-validation: 0.8891 (global LOOCV), 0.8323 (local LOOCV), and 0.8818 ± 0.0014 (5-fold CV).
• Performance metrics surpassed those of many previous computational models.
• Case studies validated the model's accuracy, with a high percentage of top-ranked miRNA predictions confirmed experimentally for esophageal neoplasms, lymphoma, lung neoplasms, and breast neoplasms.

Conclusions:

• RFMDA demonstrates high reliability and accuracy in predicting miRNA-disease associations.
• The model offers a cost-effective and efficient alternative to experimental methods.
• RFMDA has significant potential to aid in the diagnosis and treatment of complex human diseases.