AsianScientist (Oct. 15, 2020) – As the COVID-19 pandemic rages across the globe, medical professionals, scientists and public health officials are in a race against time to curb the spread of the novel coronavirus.
While we may need them now, the development of new diagnostic tools, vaccines and therapeutics remains complex and may take longer than we can wait for. This is where supercomputers, well known for their ability to rapidly process large amounts of data, can help to accelerate the process. After all, the top 500 supercomputers in the world can perform more than 1 quadrillion— that’s 1 with 15 trailing zeros—operations per second on average.
Beyond tackling the immediate challenges of infectious diseases, supercomputers can also be applied to other aspects of healthcare. From laboratory based medical research to clinical practice, here’s how supercomputers are augmenting human abilities and helping medical professionals deliver personalized care.
Discovering new therapeutics
Supercomputers that simulate how complex biological molecules behave have emerged as a useful tool in responding to the COVID-19 pandemic.
Armed with a peak performance of 1.3 petaFLOPS, the MDGRAPE-4A supercomputer at Japan’s RIKEN completed a simulation of the protease protein involved in the replication of the SARS-CoV-2 virus on March 17, 2020.
More than being just a static image, the simulation showed how the 2,416 atoms making up the protease protein move and wobble around in solution, allowing scientists to screen for potential antiviral compounds that can block it.
Such a trial-and-error process could be performed virtually as well, with supercomputers iterating through a much wider range of compounds than what is physically possible. Although not specifically applied to COVID-19 research, a software framework developed at the National Supercomputer Center in Guangzhou, China, trawled through ten million molecules in a trial run, taking just 22.31 hours using the Tianhe-2 supercomputer, which boasts a peak performance of close to 34 petaFLOPS.
Simulating virtual organs
In the past, we studied cadavers to unravel the inner workings of the human body. While gaining access to human organs is often challenging, computer simulations allow researchers to safely experiment with virtual organs that are programmed to respond in the same way.
To reproduce an organ’s complex physiology, researchers need to take into account multiple factors ranging from biochemical reactions at the subcellular level, all the way up to mechanical behavior at the organ level. Such extensive simulation can be accomplished with the help of a supercomputer.
In Japan, the T2K Open Supercomputer at the University of Tokyo was used to simulate the heart and reproduce the expected side effects of a heart problem known as arrhythmia when certain drugs were administered. Dubbed UT-Heart, the simulation can be used to screen for new medications for adverse side effects before more costly laboratory and clinical testing are conducted.
In addition, UT-Heart can even be customized based on each patient’s medical data. The team used it to predict the effectiveness of a pacemaker on each individual, paving the way for doctors to develop personalized treatment plans.
Rapid genome sequencing
In medicine, there is rarely a one-size-fit-all approach—two individuals with the same disease may show different symptoms and respond very differently to the same treatment. For example, the SARS-CoV-2 virus leads to severe pneumonia in some cases, but only a mild cough in others.
Researchers think that this is in part due to subtle variations in individual genomes, and mapping them out could suggest why certain patients are less susceptible to infections, possibly pointing the way to more effective treatments and vaccines.
Chinese company BGI Genomics, in partnership with Intel and Lenovo, has thus sought to sequence the genome of a large number of COVID-19 patients to find out why. However, genome sequencing is a timeconsuming task—the first human genome took over a decade to decode and it now typically takes about a week, even with modern technology.
To give high performance computers an extra boost, Lenovo fine-tuned both hardware and software, designing an architecture known as the Genomics Optimization and Scalability Tool (GOAST) that is able to sequence an entire genome in just 5.5 hours.
Once a vaccine has been developed, gene sequencing tools could also allow scientists to predict which subpopulation of patients it would work best in.
Reading medical images
2019 was the first time there were more elderly above 65 than young people below five. As aging populations place increasing demands on healthcare, high performance computers could be called in to help.
One possibility lies in the use of artificial intelligence (AI) to help doctors analyze medical images and make diagnoses. As a job that requires many years of training in humans, it is no easier for machines. Medical data such as those from computerized tomography (CT) scans are three-dimensional, requiring huge amounts of computational power to train an AI software to recognize them.
The process, however, can be sped up by using supercomputers. A team from The University of Tokyo in Japan used the Reedbush-H supercomputer, which has a peak performance of 800 teraFLOPS, to train an AI that could automatically detect nodules (which might be cancerous) from lung CT images.
While certain training steps are typically carried out sequentially, the team redesigned the protocol to carry them out on multiple parallel GPUs simultaneously, drastically reducing the time required to train the AI from around 105 hours to 40 hours.
Once trained, the AI could assist doctors in making a diagnosis by, for example, identifying small nodules that might have been missed.
Suggesting medical treatments
While we are still far from robot doctors taking over hospitals, there are early signs that medical AI could assist doctors by providing treatment recommendations. It may even do a better job than doctors as the AI is able to scan through more medical literature than what a human could ever read.
One key challenge lies in teaching computers how to interpret information in words, in addition to numbers. IBM’s Watson for Oncology supercomputer is one example to have emerged.
When trialed at the Manipal Comprehensive Cancer Center in India, Watson read through patient records and proposed treatment courses that were in line with a physician’s recommendation in 73 percent of 638 breast cancer cases. The results of similar trials were 83 percent in Thailand and 49 percent in South Korea.
Clearly, IBM’s supercomputer is still a work in progress, but applications for high performance computing in healthcare are likely to become more common in the future. To meet increasing demand, Taiwanese company Infortrend Technology Inc. developed a data storage system named EonStor CS to facilitate the high-speed sharing of medical files. To keep it ‘future proof,’ it was designed to be easily scaled up to accommodate a growing amount of data in a cost-effective manner.
This article was first published in the print version of Supercomputing Asia, July 2020.
Click here to subscribe to Asian Scientist Magazine in print.
———
Copyright: Asian Scientist Magazine.
Disclaimer: This article does not necessarily reflect the views of AsianScientist or its staff.
