The winds of the pandemic are changing – for better and for worse. Three viable vaccines now teeter on the brink of regulatory approval, which will pave the way for broad distribution by April or May. But until then, COVID-19 cases are skyrocketing across the U.S. and Europe, leaving much of the world with a light at the end of a treacherous tunnel. The supercomputing sector, meanwhile, is quickly pivoting to triage, with the massive COVID-19 HPC Consortium announcing a second phase specifically focusing on supporting patients over the next six months. With the hunt for therapeutics more of a focus now than ever, the Exscalate4CoV consortium is upping the ante with a behemoth 70 billion-molecule simulation of the virus aimed at finding ways to ameliorate the disease in infected patients.
The Exscalate4CoV consortium, formed in mid-February, is a public-private partnership that now includes 50 partners across seven European countries (Belgium, Germany, Italy, Poland, Spain, Sweden and Switzerland). The consortium is named for Exscalate, a “smart platform against pathogens” that is accompanied by a chemical library of half a billion molecules and which is provided by Italian pharmaceutical – and Exscalate4CoV leader – Dompé.
Three numbers are key to the new simulation: 70 billion molecules; 15 interaction sites; 60 hours. In the course of evaluating each of the molecules against each of the interaction sites, the consortium simulated over 1.07 trillion interactions in just two-and-a-half days: an average of nearly five million interactions per second.
Exscalate4CoV calls this “the most complex supercomputing simulation ever realized.” While that is a bold claim, the consortium provided a firmer comparison against simulations run on Summit earlier this year, claiming a 300-fold size increase and 500-fold speedup relative to those docking simulations.
The massive simulation used a pair of supercomputers: HPC5 (35.5 Linpack petaflops), operated by oil and gas firm Eni and standing as the most powerful publicly ranked industrial supercomputer in the world; and Marconi-100 (21.6 Linpack petaflops), operated by Italy’s Cineca.
In accordance with its policies, the results of the simulation are open-access. Exscalate4CoV took the unusual step of releasing partial results of the simulation in real-time at this link, and the full results are now available at the broader Exscalate4CoV results portal. For its part, Exscalate4CoV has been analyzing the results with AI and advanced analytics through SAS Viya to identify the molecules with the best docking scores.
Exscalate4CoV’s research has led to the identification (and current clinical testing) of one pharmaceutical (raloxifene), in addition to the publication of 12 peer-reviewed articles and massive datasets from tens of thousands of experiments.