Their joining adds four more supercomputers to the pool of supercomputers made available to researchers from all over the world
Britain and Switzerland have joined a consortium led by IBM to accelerate research to find a vaccine and cure for the corona virus. Their joining adds four more supercomputers to the pool of supercomputers made available to researchers from all over the world. There are now 20 supercomputers in the database, including the American Summit - the most powerful computer in the world (with a computing capacity of 200 petaflops), Lassen - also American and the tenth most powerful in the world, and now also the Swiss Piz Daint which is the sixth most powerful in the world (with a computing capacity of 27.15 petaflops) .
The goal of the international consortium is to harness the enormous computing capacity of the supercomputers to help and accelerate the discovery of vaccines or drug treatments to fight the corona virus. Supercomputers, which include thousands or even thousands of processors working at the same time, are able to perform calculations on a very large scale and process huge amounts of information using artificial intelligence - and thus perform in a relatively short time calculations that would take years for classical computers.
The consortium (https://covid19-hpc-consortium.org/) already includes 40 organizations, government ministries, academic institutions and organizations, among them the national laboratories of the US Department of Energy, the White House, NASA, MIT, the National Science Foundation and other companies such as Google , Microsoft, HP and Amazon.
Since mid-March, the consortium has provided 58 separate scientific studies with access to supercomputer resources, with more projects expected to take place in the near future.
The Center for UK Research and Innovation (UKRI) is a British organization working with universities, research organisations, businesses, charities and government to foster research and innovation. UKRI will make available to the consortium, among other things, ARCHER, a supercomputer with a computing capacity of 2.55 petaflops located at the University of Edinburgh. The other partner that joined is the National Center for Supercomputers in Switzerland (CSCS) will provide access to the Piz Daint supercomputer, as mentioned the sixth most powerful in the world.
The total computing power of the partners in the consortium is 500 petaflops. For comparison - Frontera, the most powerful academic supercomputer located at the University of Texas, has a computing power equal to the power of 100 thousand desktop computers.
Scientists submitting research proposals to the consortium work on three different levels that deal with the virus:
- Analysis of the protein structure of the virus using molecular simulations.
- Using machine learning to identify binding points in the virus or possible molecular candidates that may lead to the discovery of a drug that will treat the disease.
- Define risk groups and predict how the virus will spread, to help health systems deal with the virus.
- NASA scientists, for example, try to predict the spread of the virus by defining risk groups. Risk models based on their results may help identify patients best suited for vaccine and antiviral clinical trials.
- A team from the University of Michigan is using a supercomputer with AI capabilities to test whether any of the more than 1,600 existing drugs approved by the FDA can deal with COVID-19. In addition, they also test more than 5,000 experimental drugs.
Oak Ridge Lab researchers use Summit supercomputer to cut search time from years to hours with NVIDIA GPUs
Ada Sedova, a biophysics researcher at the US National Laboratory in Oak Ridge, works with the world's most powerful supercomputer from home, sometimes in her pajamas. She is looking for tiny molecules that can block the COVID-19 virus that infects humans. "I dedicate all my time to this, to save people," she described at the recent Nvidia Summit held online.
The SUMMIT supercomputer was established by IBM and consists of 9,216 IBM Power 9 processors and 27,648 NVIIDA Tesla model graphics processors. His communication is based on Mellanox's InfiniBand systems, which as we know was acquired by Nvidia, a purchase that was completed only last quarter. Her efforts are particularly rewarding: two billion tests of molecules in just 24 hours.
Sedova is looking for a ligand, an organic molecule of less than 6 atoms. The correct ligand binds to the protein of the coronavirus and prevents it from infecting healthy cells.
But finding the right ligand among all the other proteins and ligands is like looking for a needle in a haystack. Experts can spend years in labs testing all the possibilities. Even a search using Summit, ORNL's supercomputer, would take four years.
That's why Sedova and her colleagues turned to the 27,648 Nvidia-made graphics processors that populate Summit to fuel the effort.
With the help of Nvidia, the researchers transferred the code to CUDA, so that it could run on Summit and thus achieved a 2.8 times improvement in speed. Another researcher, Aron Sheinberg, tripled the speed once more when he discovered a way to use OpenMP to feed the data to Nvidia processors.
Graphics processors reduced by more than an order of magnitude the amount of time required to process a database of 1.4 billion ligands and also reduced the huge variability in results that made it difficult to schedule the tests on the supercomputer.
Imaging 2 billion compounds in 24 hours
Sedova believes that with further improvements, the team will be able to test up to two billion compounds in 24 hours. This will be the first imaging on such a scale and with high resolution.
The standard workflow for detecting protein-ligand interactions is very slow and suitable for several hundred compounds being screened on a laptop. But when it comes to hundreds of thousands of files, such a stream could disable even the largest supercomputer in the world.
This is a call to action for open source developers who strive to help accelerate scientific research.
Sedova's team is leading the effort and forming a new work stream that should yield huge numbers of tasks for Summit. Sedova consults with I/O experts on the best way to create a database of all ligands.
The next step will be to conduct an experiment of about a million compounds on 108 of the 4,608 nodes of Summit. "If it works, we can move to the big run of 1.4 billion compounds with all the nodes," she says.
Reducing research on promising molecules
If the team members succeed in the task, they will send to Memphis a list of about 9,000 of the most successful compounds to be tested in the conventional laboratory with real viruses. Or then, the researchers will be required to find the needle not in a haystack but in the amount that one swings a pitchfork.
Sedova's team began its work in January, when a leading ORNL researcher, Jeremy Smith, presented the first work that used a computer - the Summit supercomputer to research drugs to fight the coronavirus. That work is still in its infancy.
Looking to the future, Sedova has ideas for additional ways to bridge the protein and ligand bonds with conventional methods in high-performance computing.
One response
Israelis participate in planning and leading teams in Israel as well
for building supercomputers.
But we are not given access to them.
Not to work at IBM