Active learning for efficient analysis of high-throughput nanopore data.

Area of Science:

• Bioinformatics
• Computational Biology
• Genomics

Background:

• Nanopore sequencing generates vast datasets for DNA, RNA, and protein analysis.
• Machine learning is increasingly used for nanopore data, but requires extensive labeled data, incurring high labor costs.

Purpose of the Study:

• To introduce active learning to reduce labeling costs for nanopore sequencing data.
• To apply and adapt active learning strategies for complex, noisy nanopore data.

Main Methods:

• Implemented advanced active learning techniques on RNA classification (RNA-CD) and Oxford Nanopore Technologies barcode (ONT-BD) datasets.
• Introduced a bias constraint to enhance sample selection in active learning for noisy nanopore sequences.

Main Results:

• Achieved baseline performance with only 15% labeled data for RNA-CD and 50% for ONT-BD.
• Demonstrated significant reduction in labeling costs, assisting expert labeling and overcoming data labeling challenges.

Conclusions:

• Active learning effectively reduces the cost and effort of labeling high-capacity nanopore data.
• This approach holds promise for broader applications in nanopore sequence analysis.