Using deep learning for gene detection and classification in raw nanopore signals

Area of Science:

• Genomics and Bioinformatics
• Computational Biology
• Molecular Biology

Background:

• Nanopore sequencing offers advantages like speed, portability, and long reads compared to next-generation sequencing.
• Current data postprocessing (basecalling, assembly) is a significant bottleneck, delaying clinical applications.
• Identifying specific genomic regions from raw nanopore signals (squiggles) is challenging.

Purpose of the Study:

• To develop a neural network method for direct analysis of raw nanopore signals (squiggles).
• To detect and classify specific genomic regions without the need for basecalling.
• To accelerate genomic data analysis for clinical practice.

Main Methods:

• A novel neural network was designed to process raw nanopore sequencing signals.
• The method directly analyzes 'squiggles' to identify and classify genomic regions.
• The neural network can optionally direct basecalling of identified regions.

Main Results:

• The neural network successfully detects and classifies specific genomic regions directly from raw nanopore signals.
• This approach bypasses the computationally intensive basecalling step for certain analyses.
• The method enables direct analysis of raw signals or targeted basecalling of relevant sequences.

Conclusions:

• A neural network-based method can significantly streamline nanopore sequencing data analysis.
• Direct analysis of raw signals reduces postprocessing time and computational demands.
• This approach has the potential for real-time genomic analysis during sequencing runs.