NVIDIA, the Silicon Valley microprocessing giant that revolutionized computer graphics when it invented the graphics processing unit (GPU) more than two decades ago, will apply its expertise toward AI-based drug discovery and development through a partnership with GlaxoSmithKline (GSK) that will also draw upon a new supercomputer being built by NVIDIA in the U.K.
Researchers from GSK and NVIDIA will work from the pharma giant’s recently established AI hub in King’s Cross in central London. GSK opened the £10 million ($12.9 million) facility last month with an initial group of 30 scientists and engineers, with the aim of applying genetic and genomic data toward designing and developing drugs and vaccines.
Headquartered in Santa Clara, CA, NVIDIA catalyzed the growth of PC gaming and revolutionized parallel computing when it invented the GPU in 1999, and later applied GPU technology to advance AI.
Kimberly Powell, NVIDIA’s Vice President of Healthcare, told GEN that the new partnership with GSK and development of Cambridge-1 reflected a “natural evolution for Nvidia.”
“For over a decade we have partnered with the medical devices ecosystem to bring innovative diagnostic imaging, robotic surgery and patient monitoring devices to the market. We also deeply collaborate with the life sciences community to accelerate HPC [high performance computing] applications critical to drug discovery like molecular modeling and simulations,” she said.
Healthcare is perfectly suited for accelerating computing, Powell said, given the size and the scale of data generated in life sciences. NVIDIA says every facet of healthcare can benefit from its technologies, whether medical imaging, drug discovery, genomics, telehealth, patient monitoring and operational efficiency.
“NVIDIA’s technology is unique in that it is a single computing platform that is embedded in a medical device or at the edge in a workstation or server at the hospital or available on every public cloud,” Powell added. “That means we can serve applications that need to be done in real-time like emergency room triage, robotic surgery or ICU patient monitoring to assist in the detection and understanding of disease all the way to studying patient populations to predicting patient outcomes.”
High Performance Expertise
NVIDIA said it will contribute to the partnership with GSK its expertise in GPU optimization and high-performance computational pipeline development—including NVIDIA Clara Discovery™, a collection of frameworks, applications, and AI models designed to enable GPU-accelerated computational drug discovery. Clara Discovery combines accelerated computing, AI and machine learning in genomics, proteomics, microscopy, virtual screening, computational chemistry, visualization, clinical imaging, and natural language processing.
Using NVIDIA Clara Federated Learning Framework, researchers from NVIDIA and Massachusetts General Brigham Hospital partnered with 20 hospitals worldwide to develop an AI model that successfully determined whether patients who presented COVID-19 symptoms in the emergency room would need supplemental oxygen hours or days after their initial exam.
Over two weeks, the partners’ model was found to be a strong predictor of the level of oxygen required by incoming patients, having achieved a 0.94 area under the curve (AUC)—the pharmacokinetic measure of the extent of exposure to a drug and its clearance rate from the body, expressed as an integral of the curve under the plot of plasma concentration of a drug versus time after dosage (The ideal AUC is 1.0).
NVIDIA said the initiative, called EXAM (EMR CXR AI Model), represented the largest, most diverse effort in federated learning, in which partners collaborate to develop models without directly sharing sensitive clinical data with each other.
GSK has already invested in NVIDIA DGX A100™ AI infrastructure, and through the newly-announced partnership will also get to use NVIDIA’s new supercomputer Cambridge-1, which NVIDIA says is the U.K.’s most powerful supercomputer.
NVIDIA said Cambridge-1 will be used by GSK and other drug developers that include AstraZeneca—as well as other partners such as portable sequencer developer Oxford Nanopore Technologies, healthcare provider Guy’s and St Thomas’ NHS Foundation Trust, and King’s College London.
Those partners are expected to start using Cambridge-1 once it comes online by the end of this year, Powell told reporters at a virtual briefing.
Powell said the companies had not partnered in AZD1222, the front-running Phase III COVID-19 vaccine candidate that AstraZeneca is developing with the University of Oxford and a spinout company.
$52M Investment
While NVIDIA said the value of its partnership with GSK, NVIDIA did say it will invest £40 million (about $52 million) toward developing Cambridge-1, the first NVIDIA supercomputer designed and built for external research access.
“We’re building new algorithms and approaches in addition to bringing together the best minds at the intersection of medicine, genetics and artificial intelligence in the U.K.’s rich ecosystem,” stated Kim Branson, PhD, senior vice president and global head of AI and ML at GSK. “This new partnership with NVIDIA will also contribute additional computational power and state-of-the-art AI technology.”
Cambridge-1 will be housed within the AI Center of Excellence that NVIDIA said it would create back in September. That center would feature a new Arm-based supercomputer envisioned as a hub for collaborations by AI researchers, other scientists, and startup companies from across the U.K.
“The Cambridge-1 supercomputer will serve as a hub of innovation for the U.K., and further the groundbreaking work being done by the nation’s researchers in critical healthcare and drug discovery,” Jensen Huang, founder and CEO of NVIDIA, said in his keynote remarks Monday at the company’s virtual GPU Technology Conference, to be held through Friday. “Tackling the world’s most pressing challenges in healthcare requires massively powerful computing resources to harness the capabilities of AI.”
As he has since the COVID-19 pandemic began, Huang spoke from his kitchen: “”NVIDIA is a full stack computing company – we love working on extremely hard computing problems that have great impact on the world. This is right in our wheelhouse,” Huang said. “We are all-in, to advance and democratize this new form of computing for the age of AI.”
“NVIDIA is a full stack computing company – we love working on extremely hard computing problems that have great impact on the world – this is right in our wheelhouse,” Huang said. “We are all-in, to advance and democratize this new form of computing – for the age of AI.”
Investors responded to NVIDIA’s announcements and Huang’s remarks by sending the company’s shares up 5% from Friday’s closing price of $522.49, over the past two trading days, to $549.46 at the close of trading today.
Oxford Nanopore’s ‘Super’ Projects
At Oxford Nanopore, a company spokesperson told GEN that the supercomputer will make possible three kinds of projects.
The first type is training new analysis algorithms, to drive continuous performance improvement. Many of Oxford Nanopore’s performance improvements over the years have resulted from improving analysis methods rather than improving the original data generated by a nanopore device. Nanopore sequencing has seen dramatic improvements from applying AI, and as new more sophisticated methods are developed, training these algorithms becomes a process that is hungry for both processing power and data.
“We can use supercomputers like Cambridge-1 to train our algorithms on increasingly large and more diverse datasets. Researchers who have previously generated datasets on nanopore devices are able to go back to the original data and re-analyze it with more performant algorithms,”
Another category of projects enabled by Cambridge-1, according to Oxford Nanopore, will be supporting nanopore sequencing users working on large-scale discoveries, such as matching of variants and phenotypes. As customers start to generate larger and larger datasets with nanopore data, they may wish to do large-scale discovery and associating genotype/phenotype over a large number of human genomes, for example.
While Oxford Nanopore’s PromethION benchtop sequencing system includes a powerful compute module—it is the company’s highest throughput sequencer with 48 flow cells that are capable of generating up to 8 Tb of data—researchers studying really huge datasets may either use their own data centers or go elsewhere, such as Cambridge-1.
In recent months, Oxford Nanopore has applied its sequencing tech toward diagnosing COVID-19: In June, Oxford Nanopore partnered with Abu Dhabi-based Group 42 (G42) to develop a population-scale technology that rapidly and accurately detects SARS-CoV-2. G42 uses Artemis, a supercomputer that is powered by NVIDIA GPUs and is No. 26 on the latest TOP500 list of the world’s most powerful supercomputers. Two months later, Oxford Nanopore announced the rollout of LamPORE, a COVID-19 test, in an agreement with the United Kingdom’s Department of Health and Social Care.
A third type of project enabled by Cambridge-1 is biological analysis, especially of proteins. Oxford Nanopore notes that its DNA/RNA sequencing algorithms already perform a complex signal processing task during base calling, even though DNA/RNA molecules only have four canonical chemically similar bases forming the signal. In contrast, the long-sought after goal of sequencing proteins requires an approach to analyzing building blocks of over 20 amino acids.
“The computational power required to train those analysis algorithms is formidable, but potentially possible with supercomputers and machine learning,” the Oxford Nanopore spokesperson added.
Going Green
Powered by 80 NVIDIA DGX A100 systems connected by NVIDIA Mellanox© InfiniBand networking, Cambridge-1 will be an NVIDIA DGX SuperPOD™ system designed to deliver more than 400 petaflops of AI performance and 8 petaflops of Linpack performance. That would rank Cambridge-1 at No. 29 on the TOP500 list of powerful supercomputers. Cambridge-1 would also rank among the world’s top three most energy-efficient supercomputers on the current Green500 list, NVIDIA said.
DGX SuperPOD ‘s modular architecture enables the system to be installed and operational in as little as a few weeks, NVIDIA said, compared with traditional supercomputers that can take years to deploy.
NVIDIA has also applied its DGX SuperPOD to “Bio-Megatron,” a large clinical and scientific natural language processing (NLP) model developed by NVIDIA and trained with corpa of clinical and scientific corpora publications referenced in PubMed, which comprises more than 30 million citations for biomedical literature.
According to NVIDIA, Bio-Megatron can be used to extract key concepts and relations from biomedical text and build knowledge graphs for research and discovery. It can also identify clinical terms in clinical speech and text, and map them to a standardized ontology to assist in clinical documentation and research.
In a study abstract published last month and presented at the Conference on Machine Intelligence in Medical Imaging, held September 13-14, Bio-Megatron was shown to be a fast and accurate automatic speech recognition system that can capture key clinical named entities and map them to concepts in a standardized ontology. Bio-Megatron showed the highest precision and F1 score compared with a general domain Bidirectional Encoder Representations from Transformers (BERT) model, and a BERT-base model pre-trained on PubMed abstracts called BioBERT, though BioBERT scored slightly higher than Bio-Megatron in recall.
“In a time where healthcare providers and call centers are experiencing an unprecedented increase in patient call volume, we hope our contribution will help achieve faster and better patient responses, ultimately leading to improved patient care,” Hoo-Chang Shin, PhD, an NVIDIA research scientist, Christopher Parisien, PhD, of the University of Toronto, and two colleagues
Cambridge-1 will be housed within the AI Center of Excellence that NVIDIA said it would create back in September. That center would feature a new Arm-based supercomputer envisioned as a hub for collaborations by AI researchers, other scientists, and startup companies from across the U.K.
“The Cambridge-1 supercomputer will serve as a hub of innovation for the U.K., and further the groundbreaking work being done by the nation’s researchers in critical healthcare and drug discovery,” Jensen Huang, founder and CEO of NVIDIA, said in his keynote remarks today at the company’s GPU Technology Conference, to be held through Friday. “Tackling the world’s most pressing challenges in healthcare requires massively powerful computing resources to harness the capabilities of AI.”
