Ensemble-based classification approach for micro-RNA mining applied on diverse metagenomic sequences.

Area of Science:

• Bioinformatics
• Genomics
• Computational Biology

Background:

• MicroRNAs (miRNAs) are key gene expression regulators, but experimental identification is challenging due to low expression and stability.
• Computational identification of miRNAs offers a valuable alternative to traditional cloning methods.

Purpose of the Study:

• To develop and evaluate an ensemble classifier for enhanced microRNA (miRNA) prediction in genomic and metagenomic sequences.
• To improve the accuracy and efficiency of identifying novel miRNAs from diverse environmental samples.

Main Methods:

• An ensemble classifier was created by integrating four established miRNA predictors (Triplet SVM, Mipred, Virgo, EumiR) using a neural network.
• The ensemble model was trained on diverse datasets and tested on real and pseudo miRNA sequences.
• The approach was applied to metagenomic sequences from mine drainage, groundwater, and marine environments.

Main Results:

• The ensemble classifier achieved high performance metrics, including 89.3% accuracy and a 0.9 area under the ROC curve.
• It demonstrated significant performance improvements over individual classifiers like Triplet-SVM, Virgo, and EumiR.
• 179 highly probable miRNAs were identified in metagenomic samples, with some candidates validated against miRBase.

Conclusions:

• The developed computational tool effectively predicts microRNA hairpins in genomic and metagenomic data.
• The ensemble approach enhances miRNA discovery, identifying potential candidates missed by individual classifiers.
• This method facilitates the mining of metagenomic data for novel and homologous miRNAs, supporting future experimental validation.