An ensemble learning framework for potential miRNA-disease association prediction with positive-unlabeled data

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) play crucial roles in diverse disease pathogenesis.
• Accurate prediction of miRNA-disease associations is vital for understanding disease mechanisms.
• Supervised machine learning methods for miRNA-disease prediction are hindered by the lack of experimentally validated negative samples.

Purpose of the Study:

• To develop an improved computational framework for predicting miRNA-disease associations.
• To address the challenge of limited negative samples in supervised learning for miRNA-disease prediction.
• To enhance the accuracy and reliability of identifying pathogenic miRNA-disease relationships.

Main Methods:

• Proposed a novel ensemble learning framework to tackle the positive-unlabeled (PU) learning problem.
• Incorporated a semi-supervised K-means (SS-Kmeans) algorithm to extract reliable negative miRNA-disease pairs from unknown pairs.
• Utilized a subagging method for generating diverse training sets and a random vector functional link (RVFL) network for prediction.

Main Results:

• The proposed ensemble learning framework demonstrated superior prediction accuracy compared to existing popular approaches.
• The SS-Kmeans method effectively identified reliable negative samples, crucial for model training.
• A case study on lung and gastric neoplasms validated the framework's efficacy in identifying miRNA-disease associations.

Conclusions:

• The developed ensemble learning framework offers a robust solution for predicting miRNA-disease associations, overcoming limitations of traditional methods.
• The method enhances the discovery of potential miRNA biomarkers for various diseases.
• This approach provides a valuable tool for advancing research in miRNA-mediated disease mechanisms.