Mammalian microRNA prediction through a support vector machine model of sequence and structure

Area of Science:

• Genomics
• Bioinformatics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are small noncoding RNAs regulating gene expression.
• Thousands of mammalian miRNAs remain undiscovered, necessitating advanced discovery methods.
• Existing databases contain hundreds of known mouse and human miRNAs.

Purpose of the Study:

• To develop and present mirCoS, a novel computational tool for discovering new microRNA candidates.
• To leverage sequence, secondary structure, and evolutionary conservation for miRNA identification.
• To accelerate the discovery of low-expression or context-specific miRNAs.

Main Methods:

• Sequential application of three support vector machine models.
• Utilizing sequence, secondary structure, and cross-species conservation (human-mouse) as features.
• Developing a computational pipeline named mirCoS.

Main Results:

• mirCoS effectively identifies known miRNAs and predicts numerous novel hairpin structures.
• The method predicted 3476 mouse and 3441 human miRNA candidates.
• Predicted hairpins exhibit characteristics similar to known miRNAs, with some supported by expression data.

Conclusions:

• mirCoS offers a powerful approach for identifying novel microRNAs.
• The method performs comparably to or exceeds existing miRNA prediction tools.
• The study provides a substantial list of new miRNA candidates for experimental validation.