Demultiplexing barcoded Oxford Nanopore reads with deep convolutional neural networks

Abstract

Multiplexing, the simultaneous sequencing of multiple barcoded DNA samples on a single flow cell, has made Oxford Nanopore sequencing cost-effective for small genomes. However, it depends on the ability to sort the resulting sequencing reads by barcode, and current demultiplexing tools fail to classify many reads. Here we present Deepbinner, a tool for Oxford Nanopore demultiplexing that uses a deep neural network to classify reads based on the raw electrical read signal. is 'signal-space' approach allows for greater accuracy than existing 'base-space' tools (Albacore and Porechop) for which signals must first be converted to DNA base calls, itself a complex problem that can introduce noise into the barcode sequence. To assess Deepbinner and existing tools, we performed multiplex sequencing on 12 amplicons chosen for their distinguishability. is allowed us to establish a ground truth classification for each read based on internal sequence alone. Deepbinner had the lowest rate of unclassified reads (7.8%) and the highest demultiplexing precision (98.5% of classified reads were correctly assigned). It can be used alone (to maximise the number of classified reads) or in conjunction with other demultiplexers (to maximise precision and minimise false positive classifications). We also found cross-sample chimeric reads (0.3%) and evidence of barcode switching (0.3%) in our dataset, which likely arise during library preparation and may be detrimental for quantitative studies that use multiplexing. Deepbinner is open source (GPLv3) and available at https://github.com/rrwick/Deepbinner.

Solyntus, the New Highly Contiguous Reference Genome for Potato (Solanum tuberosum)

“…Initially, 6 917 092 long reads, totaling 71.9 Gb of data, were obtained using the Oxford Nanopore GridION platform. After removing adapters with Porechop ( Wick 2018 ) and filtering with Filtlong ( Wick 2019 ), 3 735 580 sequences were left with a mean length of 4 996 bps and length ranging from 4 646 – 420 405 bps and a GC content of 35% according to NanoComp ( De Coster et al 2018 ).…”

“…Quality control was performed on the long reads using minIONQC ( Lanfear et al 2019 ) and NanoComp ( De Coster et al 2018 ). Adaptors were removed using Porechop ( Wick 2018 ) and the data were filtered with Filtlong ( Wick 2019 ), removing the worst 10% of read bases while prioritizing read length.…”

The advantages and disadvantages of short- and long-read metagenomics to infer bacterial and eukaryotic community composition

“…These technologies allow read lengths of 92 10 kilobase pairs (Kbp) and beyond, in strong contrast with the approximately 300 base 93 pairs (bp) limit of Illumina. However, both PacBio and Nanopore technologies have far 94 higher error rates (88-94% accuracy for Nanopore (Wick, Judd, and Holt 2018) and 85-87% 95 for PacBio (Ardui et al 2018)). The lower accuracy of Nanopore and PacBio (non-circular 96 consensus) sequence reads may affect the success of current classification methods, and 97 there are few algorithms designed to exploit long-read data.…”

Draft Genome Sequence of Sinorhizobium meliloti AK555

“…The run yielded 1.6 million reads after basecalling (4.2 Gbp). We demultiplexed the resulting reads using Deepbiner version 0.2.0 (7) and cleaned the reads using Porechop version 0.2.3 (https://github.com/rrwick/Porechop), both with default parameters. In total, 26,889 reads with an N 50 value of 31,266 bp comprising 195 Mbp were produced for the strain AK555.…”