Semi-supervised miRNA

Semi-supervised miRNA classifies microRNA (miRNA) sequences using a semi-supervised machine learning framework that leverages labeled and unlabeled RNA sequence data, including next-generation sequencing reads, to improve miRNA identification across species.

Key Features:

• Semi-Supervised Learning: Employs a semi-supervised approach that integrates small labeled datasets with large volumes of unlabeled RNA sequence data.
• Active Learning and Multi-View Co-Training: Integrates active learning with multi-view co-training to iteratively select informative unlabeled instances and combine different feature representations.
• Next-Generation Sequencing Support: Leverages RNA sequence data generated by next-generation sequencing platforms as input for classification.
• Minimal Labeled Data Requirement: Optimized to improve classification performance with limited labeled training examples.
• High Recall and Precision: Demonstrated significant improvements in recall and precision in tests across six diverse species.

Scientific Applications:

• Novel miRNA Identification: Enables identification of novel miRNAs in newly sequenced genomes and less-studied species where known examples are scarce.
• Support for Experimental Validation: Bridges computational predictions with experimental validation workflows to facilitate miRNA discovery.

Methodology:

Initial training on available labeled data is followed by iterative active learning cycles that select the most informative unlabeled instances while multi-view co-training leverages different feature representations of the same data.