Amid rising global cases and threatening variants, a major gap remains to be filled in the world’s strategy for defeating the pandemic: effective therapeutics. Now, a Massachusetts startup powered by a wide range of HPC resources – including hometown resources from the Massachusetts Green High Performance Computing Center (MGHPCC) – has identified four promising FDA-approved drugs whose efficacy as COVID-19 therapeutics is supported by data from millions of patients.
“We developed our own simulation suite,” explained Joy Alamgir, founder of the startup in question, called ARIScience, in an interview with HPCwire. “That was the genesis of ARIScience, where we wanted a novel way to look at compounds – particularly, compounds that we want to interrupt – and a way to put in quantum observations along with classical simulation on the ligand side. … And we purposely wanted to do it homogeneously and on the same development stack, which allows us to effectively maintain it and also allows us to avoid any installation heartache that we may have as we use different kinds of compute resources that are available to us.”
A couple years in, of course, the landscape changed somewhat. “Once COVID hit the U.S. back around April of last year,” Alamgir said, “we redirected our efforts to see if there was an existing compound from all the FDA-approved drugs that we could use our simulation platform with to see if there were specific coronavirus proteins that we could interrupt.”
Alamgir wanted to simulate the structures of 1,513 FDA-approved drugs, then conduct a free energy analysis against 11 key SARS-CoV-2 proteins to see which of those drug molecules showed the best potential for disrupting those proteins.
Initially, Alamgir worked with ARIScience’s in-house HPC cluster – a small set of “essentially three nodes” equipped with Intel CPUs dating back a couple generations of hardware.
“Our internal HPC very quickly ran out of computing power,” Alamgir said, “at which point we reached out to [John Goodhue, executive director of the] MGHPCC and a few other organizations to see if they could allow us to run it on their HPC platforms.” Beyond the MGHPCC, Alamgir received allocations at the Pittsburgh Supercomputing Center (PSC), the Texas Advanced Computing Center (TACC) and the Universities of Maine and North Dakota.
None of the allocations were massive, and Alamgir took advantage of the flexibility of ARIScience’s platform to spread the load across the systems using Slurm. On each system, he ballparked, the project was using somewhere between six and 30 nodes at a time, with each simulation job taking a few hours and all 1,513 analyses for one protein taking several days to complete.
At the end of all that simulating, Alamgir was left with 18 promising FDA-approved drugs.
“At which point we were like, okay, great, we have these 18 results,” he said. “What do we do with them, right? It’s not like we can run 18 clinical trials.”
So Alamgir decided to take the analysis one step further and vet the results against real patient data. After “a lot of requesting,” ARIScience was granted access to data from the National COVID Cohort Collaborative (N3C) data from the National Institutes of Health (NIH), making the company one of the first commercial entities to gain access to the massive dataset, which contains detailed – and yes, deidentified – data from 1.5 million patients.
Carefully controlling for demographic data that might affect the results, Alamgir then used a “very sophisticated statistical analysis” to examine the real-world differences in mortality between patients who used one of those 18 drugs and patients who didn’t.
The result: four drugs (amoxicillin, metformin, hydrochlorothiazide and triamcinolone) that were each “statistically significantly associated with reduced COVID mortality of about 25 percent.” Furthermore – although the sample size was lower and the data analysis remains ongoing – Alamgir shared that the combination effect of hydrochlorothiazide and metformin appeared to be even stronger. “The mortality rate reduction was the highest for the patients that were taking both of them,” he said. “We detected a 41 percent reduction in COVID mortality odds.” (The other combination effects had not yet been similarly assessed.)
For now, Alamgir’s analysis is constrained to mortality – the most crucial outcome – and further patient data may be a bit more difficult to work with. “We purposefully froze our analysis using a data release of about mid-December of 2020,” he said. “The reason primarily is that starting in late December, the vaccination campaigns in the U.S. have started and depending on who got vaccinated or not, you introduce additional unknowns into this statistical analysis.”
So now, Alamgir and ARIScience are running with what they have: highly promising results for four FDA-approved drugs. With vaccines taking hold in the U.S. and critical peaks occurring elsewhere around the world, the company is turning its attention to South America and South Asia to explore the possibility of a randomized clinical trial to further bolster the drugs’ efficacy in staving off COVID mortality.
For Goodhue and the MGHPCC, this is a familiar success story, with the director characterizing the partnership between the MGHPCC and ARIScience as one of the center’s many actions “at the trailing edge of research and the leading edge of societal impact.” The center’s previous accounts, he said, included the founders of Moderna and a range of successful startups. “All of them started with us when they were early-stage,” he said. “If I was trying to impress people, I’d say we’ve managed to help a couple of companies on their way to their exits by acquisition, I think yielding a total of three quarters of a billion dollars.”
During the pandemic, the MGHPCC has been running what Goodhue explains as a “mini version” of the COVID-19 HPC Consortium, offering computing resources to companies based in Massachusetts with research ideas for tackling the virus. Right now, Goodhue said, the center is actively partnered with around a half dozen companies.