Mirnacle: machine learning with SMOTE and random forest for improving selectivity in pre-miRNA ab initio prediction

Area of Science:

• Molecular Biology
• Bioinformatics
• Genomics

Background:

• MicroRNAs (miRNAs) are vital gene regulators in biological processes.
• In silico prediction of miRNA precursors (pre-miRNAs) is crucial but faces challenges with false positives.
• Current ab initio methods for miRNA prediction lack sufficient selectivity.

Purpose of the Study:

• To improve the selectivity of ab initio pre-miRNA prediction.
• To reduce the number of false positives in computational miRNA identification.
• To enhance the efficiency of molecular biology research involving miRNAs.

Main Methods:

• An extension of the miRNAFold method was developed, named Mirnacle.
• Machine learning techniques, specifically random forest, were employed.
• The SMOTE procedure was utilized to address imbalanced datasets.

Main Results:

• Mirnacle demonstrated substantial improvements in selectivity without compromising sensitivity.
• The method achieved at least 97% sensitivity across three experimental datasets.
• Selectivity increases of two-fold, 20-fold, and six-fold were observed compared to existing tools.

Conclusions:

• The integration of machine learning significantly enhances pre-miRNA prediction selectivity.
• Mirnacle reduces the burden of experimental validation for miRNA studies.
• This computational tool is expected to benefit research on miRNAs, including those related to diseases.