Reliable prediction of Drosha processing sites improves microRNA gene prediction

Area of Science:

• Genomics
• Molecular Biology
• Bioinformatics

Background:

• Mature microRNAs (miRNAs) are derived from hairpin precursors through enzymatic processing.
• The initial Drosha processing step is crucial as it defines the mature miRNA product and is conserved across miRNA genes.
• Accurate identification of Drosha processing sites is vital for discovering novel miRNA genes.

Purpose of the Study:

• To develop a computational method for predicting Drosha processing sites in candidate miRNA hairpins.
• To improve the accuracy and efficiency of miRNA gene discovery.
• To re-evaluate existing miRNA annotations based on processing site prediction.

Main Methods:

• Development of a Support Vector Machine (SVM) classifier, termed Microprocessor SVM, to predict 5' Drosha processing sites.
• Training a secondary classifier on the output of Microprocessor SVM to enhance prediction of unconserved miRNAs.
• Reanalysis of miRNA characteristics and supporting evidence for newly annotated miRNAs.

Main Results:

• Microprocessor SVM correctly predicts the 5' Drosha processing site for 50% of known human 5' miRNAs.
• 90% of Microprocessor SVM predictions are within two nucleotides of the true processing site.
• A secondary classifier trained on Microprocessor SVM output outperforms existing methods for predicting unconserved miRNAs.

Conclusions:

• The developed classifiers provide a robust method for identifying miRNA processing sites and discovering new miRNA genes.
• Some previously annotated miRNAs may be misannotated, highlighting the need for experimental validation as Drosha and Dicer substrates.
• The computational tools are publicly available for use in miRNA research.