miRNA-dis: microRNA precursor identification based on distance structure status pairs

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• MicroRNA precursor identification is crucial in bioinformatics.
• Existing Support Vector Machine (SVM) methods have limited feature vector discriminative power and lack interpretable models.
• Prior studies suggest sequence and structure order effects are relevant but underexplored for pre-microRNA identification.

Purpose of the Study:

• To propose a novel method, miRNA-dis, for enhanced microRNA precursor identification.
• To incorporate structure-order information into prediction models.
• To develop an interpretable model for analyzing characteristic sequence features.

Main Methods:

• Constructed feature vectors using the occurrence frequency of "distance structure status pairs" or "distance-pairs".
• Employed rigorous cross-validations on a large, stringent benchmark dataset.
• Trained the model on human pre-microRNA data.

Main Results:

• miRNA-dis outperformed state-of-the-art predictors.
• Achieved 87.02% accuracy in predicting 4022 pre-miRNAs across 11 diverse species (animals, plants, viruses).
• The model demonstrated interpretability, revealing potential microRNA characteristics.

Conclusions:

• miRNA-dis is a powerful tool for high-throughput microRNA precursor analysis.
• The method offers improved accuracy and interpretability compared to existing approaches.
• A publicly accessible web server facilitates widespread use of miRNA-dis.