Predicting potential miRNA-disease associations based on more reliable negative sample selection.

Area of Science:

• Biomedical informatics
• Computational biology
• Machine learning in genomics

Background:

• MicroRNA (miRNA) dysfunction is linked to numerous human diseases.
• Accurate identification of disease-associated miRNAs is crucial for understanding disease mechanisms.
• Supervised learning methods for miRNA-disease association prediction require reliable negative samples, which are often unavailable.

Purpose of the Study:

• To develop a computational method for selecting more reliable negative samples for miRNA-disease association predictions.
• To improve the performance of supervised learning models by addressing the challenge of negative sample selection.

Main Methods:

• Proposed KR-NSSM method integrating two semi-supervised algorithms.
• Utilized a refined K-means algorithm for initial screening of miRNA-disease samples.
• Employed a Rocchio classification-based method for further refinement of negative and positive samples.

Main Results:

• KR-NSSM effectively selects more reliable negative samples compared to random selection.
• Ablation tests confirmed the benefit of combining K-means and Rocchio classification.
• Prediction accuracy improved across six classifiers and five prediction models using KR-NSSM-selected samples.
• 469 out of 1123 selected positive miRNA-disease associations were validated by existing databases.

Conclusions:

• KR-NSSM significantly enhances the performance of supervised machine learning in miRNA-disease association prediction.
• The method provides a valuable tool for negative sample selection in biomedical research.
• Reliable negative sample selection is critical for advancing miRNA-disease association studies.