A semi-supervised machine learning framework for microRNA classification.

Area of Science:

• Bioinformatics
• Genomics
• Molecular Biology

Background:

• MicroRNAs (miRNAs) are key regulators of gene expression, essential for cellular activities.
• Identifying miRNAs is resource-intensive, requiring computational and wet-lab validation.
• Supervised machine learning methods for miRNA identification demand large labeled datasets, often unavailable for less-sequenced species.

Purpose of the Study:

• To explore semi-supervised machine learning for miRNA classification, maximizing the use of both labeled and unlabeled data.
• To present a novel pipeline combining active learning and multi-view co-training for enhanced miRNA identification.

Main Methods:

• Application of a multi-stage semi-supervised machine learning pipeline.
• Integration of active learning and multi-view co-training techniques.
• Testing across six diverse species to evaluate performance.

Main Results:

• The proposed semi-supervised approach significantly improves miRNA classification performance.
• The method demonstrates high efficacy even with very small numbers of labeled instances.
• Effective utilization of available unlabeled RNA sequence data was achieved.

Conclusions:

• The semi-supervised miRNA classification pipeline can identify novel miRNAs with high recall and precision.
• The method requires minimal previously known miRNA data, beneficial for niche species.
• This approach offers a valuable tool for studying miRNAs in newly sequenced genomes.