Feature Selection Has a Large Impact on One-Class Classification Accuracy for MicroRNAs in Plants

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNAs (miRNAs) are crucial regulators of gene expression.
• Experimental miRNA analysis is complex, necessitating computational detection methods.
• Current computational approaches predominantly use two-class machine learning, facing challenges with limited negative training data.

Purpose of the Study:

• To investigate the efficacy of one-class classification for computational miRNA detection.
• To evaluate the impact of feature selection methods on one-class classification accuracy for plant miRNAs.
• To improve upon existing two-class classification methods for miRNA identification.

Main Methods:

• Employed feature selection procedures in conjunction with one-class classification algorithms.
• Utilized over 700 described features for miRNA parametrization.
• Compared the accuracy of various feature selection methods for training one-class classifiers.

Main Results:

• Feature selection methods demonstrated up to a 36% difference in accuracy.
• The optimal feature set enabled a one-class classifier to achieve an average accuracy of approximately 95.6%.
• This one-class approach slightly outperformed previous two-class-based plant miRNA detection methods by ~0.5%.

Conclusions:

• One-class classification, combined with effective feature selection, is a highly accurate method for computational miRNA detection.
• Future improvements can be achieved through rigorous filtering of positive training data and enhanced feature clustering algorithms.
• This study provides a more robust and efficient alternative for plant miRNA identification.