NanoReviser

NanoReviser corrects basecalling errors in Oxford Nanopore Technologies (ONT) nanopore sequencing reads using deep learning to improve sequencing accuracy for genomic and epigenetic analyses.

Key Features:

• Deep learning models: Uses convolutional neural networks (CNNs) and bidirectional long short-term memory (Bi-LSTM) networks that leverage raw electrical signals and basecalled sequences to refine basecalls.
• Re-segmentation of raw signals: Re-segments raw ONT electrical signals based on initial basecalls to enable more precise correction.
• Post-basecalling correction without consensus building: Performs error correction as a post-basecalling reviser without requiring time‑intensive consensus-sequence construction.
• Empirical error reduction: Demonstrated over 5% reduction in overall error rate on public E. coli and human NA12878 ONT datasets.
• Methylation-aware training: Incorporating methylation information during training reduced the error rate by 7% overall and by more than 10% in methylation-rich regions on E. coli data.

Scientific Applications:

• Microbial genomics: Improves long-read accuracy for microbial genome assembly and variant detection using ONT data.
• Human genetic studies: Enhances basecalling quality for human NA12878 and other human ONT sequencing projects to support variant analysis.
• Epigenetic analysis (methylation): Supports analysis of methylation patterns by reducing basecalling errors in methylation-rich regions when methylation information is used in training.

Methodology:

Post-basecalling correction using CNN and Bi-LSTM models that leverage both raw ONT electrical signals and basecalled sequences, with re-segmentation of raw signals based on initial basecalls and optional incorporation of methylation information into training.