A clustering-based sampling method for miRNA-disease association prediction.

Area of Science:

• Computational biology
• Genomics
• Biomedical informatics

Background:

• MicroRNAs (miRNAs) are crucial regulators of gene expression, and their dysregulation is linked to complex human diseases.
• Experimental identification of disease-associated miRNAs is costly and inefficient, necessitating advanced computational approaches.
• Existing computational methods face challenges in constructing reliable negative sample sets for accurate prediction.

Purpose of the Study:

• To develop and validate a novel computational method, CSMDA, for predicting miRNA-disease associations.
• To address limitations in negative sample set construction prevalent in current prediction models.
• To improve the accuracy and efficiency of identifying potential disease-associated miRNAs.

Main Methods:

• Integrated multiple miRNA and disease similarity information to represent miRNA-disease pairs.
• Implemented a clustering-based sampling method for constructing a robust negative sample set, avoiding potential positive sample inclusion.
• Utilized random forest for feature selection to reduce noise and redundancy in high-dimensional data.
• Employed an ensemble learning framework with soft voting for final miRNA-disease association prediction.

Main Results:

• CSMDA achieved high performance metrics: Precision (0.9676), Recall (0.9545), F1-score (0.9610), AUROC (0.9928), and AUPR (0.9940) via five-fold cross-validation.
• Case studies on three cancers demonstrated that the top predicted miRNA-disease associations were validated by existing databases and literature.
• The proposed method significantly improves the accuracy of predicting potential disease-associated miRNAs.

Conclusions:

• CSMDA offers a more accurate and efficient computational approach for predicting miRNA-disease associations.
• The novel negative sampling strategy and ensemble learning framework contribute to the method's superior performance.
• This approach has the potential to accelerate the discovery of novel biomarkers and therapeutic targets for various diseases.