Genome-wide pre-miRNA discovery from few labeled examples

Area of Science:

• Bioinformatics
• Computational Biology
• Machine Learning

Background:

• Supervised learning methods for microRNA hairpin prediction face limitations due to scarce positive examples, difficulty in generating negative examples, and severe class imbalance.
• These challenges hinder accurate and efficient genome-wide prediction of novel microRNAs.

Purpose of the Study:

• To develop an efficient and accurate method for genome-wide prediction of novel microRNAs.
• To address the limitations of supervised learning in microRNA hairpin identification.

Main Methods:

• Proposed miRNAss, a novel semi-supervised learning method for microRNA hairpin prediction.
• Incorporated unlabeled stem-loop sequences to enhance prediction rates.
• Developed an automatic method for generating negative examples to aid algorithm initialization.

Main Results:

• MiRNAss achieved superior prediction rates and reduced execution times compared to state-of-the-art supervised methods.
• The method demonstrated effectiveness even with limited and unrepresentative labeled data.
• Validated on genome-wide data from three model species, processing over one million hairpin sequences each.

Conclusions:

• MiRNAss offers an efficient and accurate solution for genome-wide microRNA hairpin prediction.
• The semi-supervised approach effectively leverages unlabeled data, overcoming limitations of traditional supervised methods.
• The method is applicable to real-world prediction tasks and publicly available as an R package and web demo.