EGBMMDA: Extreme Gradient Boosting Machine for MiRNA-Disease Association prediction

Area of Science:

• Genomics
• Bioinformatics
• Computational Biology

Background:

• Identifying associations between microRNAs (miRNAs) and human diseases is crucial for medical research.
• Computational inference of potential miRNA-disease associations improves experimental efficiency by prioritizing candidates for verification.
• Predicting miRNA-disease associations is a rapidly growing area of importance in biomedical research.

Purpose of the Study:

• To present a novel computational model, Extreme Gradient Boosting Machine for MiRNA-Disease Association (EGBMMDA), for predicting potential miRNA-disease associations.
• To integrate miRNA functional similarity, disease semantic similarity, and known miRNA-disease associations into a predictive framework.

Main Methods:

• Developed EGBMMDA, a model utilizing Extreme Gradient Boosting Machine (a decision tree learning method).
• Constructed feature vectors for miRNA-disease pairs using statistical, graph theoretical, and matrix factorization measures.
• Trained the model on known miRNA-disease associations from the HMDD v2.0 database.

Main Results:

• EGBMMDA achieved reliable performance with AUCs of 0.9123 (global) and 0.8221 (local) in leave-one-out cross-validation.
• The model demonstrated stability with an AUC of 0.9048 ± 0.0012 in fivefold cross-validation.
• Case studies showed high experimental validation rates for top predictions: 98% for Colon Neoplasms, 90% for Lymphoma, 98% for Prostate Neoplasms, 100% for Breast Neoplasms, and 98% for Esophageal Neoplasms.

Conclusions:

• EGBMMDA is the first decision tree learning-based model for predicting miRNA-disease associations.
• The model demonstrates robust performance and stability, making it a valuable computational resource.
• EGBMMDA effectively predicts potential miRNA-disease associations, aiding in experimental research and discovery.