SquiggleNet: real-time, direct classification of nanopore signals

Abstract

We present SquiggleNet, the first deep-learning model that can classify nanopore reads directly from their electrical signals. SquiggleNet operates faster than DNA passes through the pore, allowing real-time classification and read ejection. Using 1 s of sequencing data, the classifier achieves significantly higher accuracy than base calling followed by sequence alignment. Our approach is also faster and requires an order of magnitude less memory than alignment-based approaches. SquiggleNet distinguished human from bacterial DNA with over 90% accuracy, generalized to unseen bacterial species in a human respiratory meta genome sample, and accurately classified sequences containing human long interspersed repeat elements.

Rapid PCR-Based Nanopore Adaptive Sequencing Improves Sensitivity and Timeliness of Viral Clinical Detection and Genome Surveillance

“…The existing adaptive sequencing methods are primarily based on the alignment of nucleotide sequences or raw electrical signals, which could achieve approximately a 5-fold maximum enrichment of target genome data ( Gan et al, 2021 ; Kipp et al, 2021 ; Kovaka et al, 2021 ; Payne et al, 2021 ; Wanner et al, 2021 ; Martin et al, 2022 ). A deep-learning model distinguishes human DNA from bacterial DNA with over 90% accuracy and is faster than alignment-based approaches, which could represent an alternative option for adaptive sequencing to increase enrichment efficiency ( Bao et al, 2021 ).…”

“…Oxford Nanopore Technologies recently launched an adaptive sequencing function by aligning the reads to the references and ejecting uninterested reads by reversing the voltage across individually selected nanopores in real-time, which could achieve a computational enrichment of on-target sequences without an additional pretreatment process ( Loose et al, 2016 ; Bao et al, 2021 ). Previous studies have developed adaptive sequencing tools using a graphical processing unit base-calling (readfish) ( Payne et al, 2021 ) or raw electrical signal mapping (UNCALLED) ( Kovaka et al, 2021 ).…”

The Application of Metagenomics to Study Microbial Communities and Develop Desirable Traits in Fermented Foods

“…As sequencing technologies are becoming more reliable, accessible, with higher throughputs and reduced costs, many food companies and regulatory bodies have moved away from culture-based and classical sequencing methods such as single nucleotide polymorphism (SNP) and multilocus sequence typing (MLST), and have generally adopted NGS alternatives [ 176 ]. The rapid analysis speeds further supported by real-time base calling and identification of microbial species, offered by third generation sequencing technologies such as ONT, allow food industries and regulatory bodies to make quick, informed decisions that are crucial to preventing and/or limiting foodborne outbreaks and bacteriophage invasions within the processing facilities [ 177 , 178 , 179 ]. Recently developed technologies such as “Read Until” in ONT platforms allow selective sequencing through the classification of the short prefix sequence of a DNA or RNA strand entering a nanopore into a target or non-target sequence.…”

Human Retrotransposons and Effective Computational Detection Methods for Next-Generation Sequencing Data

“…Contrary to the prior sequencing technologies that rely on PCR methods to amplify a given template, the third-generation sequencers have two distinctive features that can ameliorate biases resulting from the PCR procedure. First, they allow for analysis in real time, and second, they interrogate a single molecule of DNA with no need for synchronization [ 102 , 104 , 105 ]. As the read length of the third-generation sequencers is much longer than that of second-generation sequencing technologies with maximal lengths of 30–150 kb, it is expected to be established as a more applicable method to detect various SVs, especially derived from retrotransposons [ 106 , 107 , 108 , 109 ].…”