Oxford Nanopore, NVIDIA Partner on GPU-Powered PromethION
By Bio-IT World Staff
January 15, 2021 | Oxford Nanopore Technologies and NVIDIA announced a collaboration this week to integrate the NVIDIA DGX Station A100 into Oxford Nanopore’s ultra-high throughput sequencing system, PromethION. Partnering the NVIDIA A100 Tensor Core GPU technology with the PromethION device aims to deliver the world’s most powerful sequencer that supports real-time analyses at scale and can also analyze any length fragment of DNA/RNA.
The use of accelerated computing and artificial intelligence to quickly and accurately sequence DNA or RNA supports the increasing availability of nanopore sequencing data, at scale, to a variety of high-throughput users, according to Oxford Nanopore.
NVIDIA DGX Station A100, announced in November, is a data-center-grade, GPU-powered, multi-user workgroup appliance that can tackle the most complex AI workloads. It plugs directly into an outlet in an office or laboratory and is very quiet thanks to its refrigerant-based cooling system. It contains four NVIDIA A100 80GB GPUs, fully connected via NVIDIA NVLink, to offer a total of 320GB of GPU memory.
The 2.5 petaFLOP AI compute system from NVIDIA offers unprecedented compute density, performance and flexibility in a benchtop format.
Kimberly Powell, NVIDIA’s vice president of healthcare, sees software-defined instruments — devices that can be regularly updated to reflect the latest scientific understanding and AI algorithms — as key to connecting the latest research breakthroughs with the practice of medicine, according to the NVIDIA blog posted after her J.P. Morgan Healthcare conference presentation this week. Oxford Nanopore excels at making “software-defined” instruments, said Powell.
Oxford Nanopore’s PromethION P48 sequencing device continues to challenge even the most powerful devices with its ability to generate as much as 10 Terabases of DNA information per 72-hour run (sufficient to analyze 96 human genomes at 30X coverage).
Breaking through the 10 Terabase run barrier was announced by Oxford Nanopore at the Community meeting in December 2020 and represents a 25% increase in data output compared to its previous best earlier in the same year. This increase has been driven by continual improvements in flow cell chemistry, many of which were included in new shipments from mid-November 2020. These developments have been reflected in customer data, with increasing yields reported across a range of sequencing applications.
Supplied in a P24 and a P48 format, PromethION is increasingly being deployed into high-throughput projects, where the rich sequencing data provided by Oxford Nanopore can be delivered at very high throughput. As with all Oxford Nanopore devices, the technology enables academic groups, core facilities and service providers to realize the value of sequencing any length fragments, from short fragments to those that are over 100,000 bases long, and to characterize base modifications, coupled with high accuracy single nucleotide or structural variant calling and phasing.
NVIDIA GPUs are already used in other Oxford Nanopore sequencing systems, driving real-time sequencing analysis at any scale. The desktop GridION includes NVIDIA V100 technology and the handheld MinION Mk1C sequencer is powered by the NVIDIA Jetson Edge AI platform.
The use of powerful AI systems is also driving substantial improvements in the accuracy of Oxford Nanopore’s sequencing data; updated analysis algorithms can result in higher accuracy of the same sequence data.
Oxford Nanopore recently released a new machine-learning driven analysis algorithm, Bonito CRF, with which users have reported >98% single read basecalling accuracy. This latest update to Bonito builds on previous work to deliver improved performance, and is trained with a larger, more diverse data set.
This technology “truly touches on the entire practice of medicine,” Powell said of Oxford Nanopore’s work, whether COVID epidemiology or in human genetics and long read sequencing. “Through deep learning, their base calling model is able to reach an overall accuracy of 98.3 percent, and AI-driven single nucleotide variant calling gets them to 99.9 percent accuracy.”
High single read accuracy supports very high consensus accuracy (sequencing multiple times for higher accuracy); optimized analysis tools including Guppy/Bonito basecalling, assembly with Canu/Flye and polishing with Medaka, can now enable Q45 with R9.4.1 flow cells and Q50 with R10.3 flow cells.
Variant calling performance is also improving with the latest releases. Using the latest tools, structural variation (SV) accuracy has reached gold standard of 96% with 30X rather than 60X coverage. Oxford Nanopore has now seen SNV at 99.92%, which is comparable to traditional SBS accuracy.
In addition, in late 2020 Oxford Nanopore generated modal single-read accuracy of 99.1% (99%=Q20) using a new chemistry with Bonito, delivered on internal validation sets, with a substantial proportion of these raw reads above Q20, according to the company.
Oxford Nanopore and NVIDIA are working closely to deploy the latest advancements in AI, with the goal of making biological analysis available to anyone, anywhere.