DeepNano-blitz: A Fast Base Caller for MinION Nanopore Sequencers

Boža, Vladimír; Perešíni, Peter; Brejová, Broňa; Vinař, Tomáš

doi:10.1101/2020.02.11.944223

Cited by 16 publications

(26 citation statements)

References 9 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Our software can readily be adapted to work with the output of other neural network basecallers. Application to the recent DeepNano-blitz [ 11 ], showed a similar gain in accuracy from consensus decoding. We also applied our algorithm to the ONT basecaller Bonito [ 12 ], a research basecaller inspired by recent successes of purely convolutional neural networks in speech recognition, and compared results with Guppy, an earlier ONT basecaller which can make use of 1D 2 .…”

Section: Main Textmentioning

confidence: 89%

Pair consensus decoding improves accuracy of neural network basecallers for nanopore sequencing

Silvestre-Ryan

Holmes

2021

Genome Biol

View full text Add to dashboard Cite

We develop a general computational approach for improving the accuracy of basecalling with Oxford Nanopore’s 1D2 and related sequencing protocols. Our software PoreOver (https://github.com/jordisr/poreover) finds the consensus of two neural networks by aligning their probability profiles, and is compatible with multiple nanopore basecallers. When applied to the recently-released Bonito basecaller, our method reduces the median sequencing error by more than half.

show abstract

Section: Main Textmentioning

confidence: 89%

Pair consensus decoding improves accuracy of neural network basecallers for nanopore sequencing

Silvestre-Ryan

Holmes

2021

Genome Biol

View full text Add to dashboard Cite

show abstract

“…We deployed a Gated Linear Units (GLU) [30] and a fully-connected layer before and after the convolution module respectively and the kernel sizes are [3,5,7,31×3] for the overall six encoder blocks.…”

Section: Methodsmentioning

confidence: 99%

“…Therefore, it is not a trivial work to capture the complicated correlations between each signal and its surroundings. Here, we employed a convolution-augmented transformer architecture instead of opting for the already adopted RNN [7,16] or CNN [6] approaches. As shown in Fig.1, the CATCaller model has two normal convolution layers and N encoder blocks followed by a fully-connected layer to produce the predicted probabilities of each base.…”

Section: Model Architecturementioning

confidence: 99%

“…In the MinION sequencer, each signal value is determined by the multiple nucleotides reside in the nanopore (generally 5 for R9.4 pore), resulting in a large number of base combinations (4 5 =1024) and making it difficult to precisely translate the raw signals into nucleotide sequences. Additionally, a successful MinION run can produce approximately 1.5-2.0 million electrical signals per second [7], magnitude faster than the speed of most existing basecalling tools. In a struggle to keep up with the data generation speed of ONT sequencer, basecallers of fast versions have to sacrifice accuracy performance.…”

Section: Introductionmentioning

confidence: 99%

“…Chiron[18] that combined convolution neural network (CNN) and RNN together with a CTC decoder runs quite slow and thus can only be used on small datasets. More recently, three models, MinCall[6], DeepNano-blitz[7] and SACall[19], were proposed. MinCall used CNN instead of RNN to improve parallelism, however, it was released without well-trained models and trained based on data for R9 chemistry which is currently replaced by R9.4.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

An End-to-end Oxford Nanopore Basecaller Using Convolution-augmented Transformer

Chen

et al. 2020

Preprint

View full text Add to dashboard Cite

Oxford Nanopore sequencing is fastly becoming an active field in genomics, and it’s critical to basecall nucleotide sequences from the complex electrical signals. Many efforts have been devoted to developing new basecalling tools over the years. However, the basecalled reads still suffer from a high error rate and slow speed. Here, we developed an open-source basecalling method, CATCaller, by simultaneously capturing global context through Attention and modeling local dependencies through dynamic convolution. The method was shown to consistently outper-form the ONT default basecaller Albacore, Guppy, and a recently developed attention-based method SACall in read accuracy. More importantly, our method is fast through a heterogeneously computational model to integrate both CPUs and GPUs. When compared to SACall, the method is nearly 4 times faster on a single GPU, and is highly scalable in parallelization with a further speedup of 3.3 on a four-GPU node.

show abstract