Improved Pre-miRNA Classification by Reducing the Effect of Class Imbalance

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators in animal and plant biological processes.
• Existing machine learning methods for miRNA prediction struggle with imbalanced datasets of real and pseudo pre-miRNAs.
• This imbalance hinders accurate identification of species-specific miRNAs.

Purpose of the Study:

• To develop a robust pre-miRNA classification method that overcomes class imbalance issues.
• To improve the accuracy and efficiency of identifying novel and species-specific pre-miRNAs.
• To enhance the discovery of new pre-miRNAs using advanced machine learning techniques.

Main Methods:

• Proposing MiRNAClassify, a cost-sensitive ensemble learning approach.
• Employing iterative training with balanced positive and negative samples to mitigate class imbalance.
• Assigning higher cost weights to positive samples to emphasize their importance.

Main Results:

• MiRNAClassify significantly outperforms existing state-of-the-art methods and models in cross-validation tests.
• The method demonstrates superior performance on human, animal, and plant pre-miRNA datasets.
• MiRNAClassify effectively identifies novel pre-miRNAs, showcasing its discovery potential.

Conclusions:

• MiRNAClassify offers a significant advancement in pre-miRNA prediction by effectively handling class imbalance.
• The proposed method provides a reliable tool for identifying both known and novel pre-miRNAs across diverse species.
• This approach enhances the discovery of new pre-miRNAs, contributing to a deeper understanding of miRNA functions.