Scientists Use AI To Find Optimal Coronavirus Treatments
Dean Ho, director of Singapore’s N.1 Institute for Health, has used machine learning to discover the optimal combination of drugs to treat the coronavirus. Here’s how he did it
The words hydroxychloroquine and azithromycin have been splashed across almost every major news site over the last week, following US president Donald Trump’s announcement that he is taking both drugs to ward off the coronavirus—despite there being no evidence that the drugs are effective.
Hydroxychloroquine treats both malaria and arthritis, while azithromycin is an antibiotic, intended to prevent infections. Though both drugs have been approved for human use, neither have been proven as a viable treatment for the coronavirus.
Public interest in the drugs spiked soon after Trump’s televised address, sparking health officials and scientists to warn that the drugs may be unsafe and shouldn’t be taken at home, emphasising that trials for hydroxychloroquine showed potential fatal heart complications.
Dean Ho, the director of N.1 Institute for Health, an infectious diseases programme within the National University of Singapore, agrees that both drugs are relatively ineffective in the fight against the coronavirus. “Data suggests that hydroxychloroquine may be linked to cardiac issues and at this time, it hasn’t been shown to improve outcomes for patients,” says Ho, who is also head of the university's department of Biomedical Engineering.
Instead, Ho believes the optimal treatment for the coronavirus is a combination of three drugs: lopinavir and ritonavir, which are currently used to treat HIV, and remdesivir, an antiviral drug originally used to treat ebola.
Ho and his multidisciplinary team at the N.1 Institute for Health determined the optimal coronavirus treatment using their artificial intelligence platform IDentif.AI, which uses machine learning to identify optimal infectious disease therapies. In a matter of days, IDentif.AI was able to test and identify trillions of different drug doses and combinations, providing clinicians with a list of drug treatments, ranging from most optimal to least optimal. It’s important to provide clinicians with a range of drug treatments, says Ho, as it allows them to counter any potential drug supply shortages, financial restrictions or patient allergies.
The AI testing was carried out on a live strain of the coronavirus, taken from a patient in Singapore, and was studied alongside 12 clinically actionable drugs on a kidney cell—a cell type often used to study viral infections. IDentif.AI found that the combination of lopinavir, ritonavir and remdesivir had an inhibition percentage of virtually 100 percent on average—indicating that the coronavirus was completely neutralised, while the combination of hydroxychloroquine and azithromycin was relatively ineffective, at just 3 percent inhibition.
The HIV drug combination is already being considered as a viable treatment for the coronavirus by scientists around the world, but Ho believes that only when the two are paired with remdesivir does it exponentially grow in its ability to inhibit the virus.
“Our data reinforces that the combination of lopinavir and ritonavir doesn’t appear to be effective at inhibiting the coronavirus infection, but when it’s paired with remdesivir the inhibition hits pretty much close to 100 percent,” says Ho.
“Remdesivir is the most effective monotherapy,” he says. “But the drug is actually six-and-a-half times less effective on its own compared to this combination.”
A study published in the New England Journal of Medicine also showed that remdesivir was effective in shortening the course of coronavirus, but that alone, remdesivir is not a sufficient treatment. Instead, it suggests that “future strategies should evaluate antiviral agents in combination with other therapeutic approaches or combinations of antiviral agents to continue to improve patient outcomes in Covid-19.”
The testing also analysed the toxicity levels of drug combinations, determined using a human heart cell, a human liver cell and a monkey kidney cell. In other studies, all three drugs in the combination have been shown to cause low levels of liver toxicity when used in isolation; Ho’s testing highlights that there’s no increase in toxicity levels when the three drugs are taken together. “I’d say it is within acceptable levels. Because when taking any drug, even panadol, your liver will feel it very temporarily,” Ho says.
While Ho’s drug combination looks promising in its ability to inhibit the coronavirus, there are still a few bridges to cross. Having recently published their findings on the online medical manuscript platform medRxiv, Ho is now waiting for his study to be peer-reviewed and approval to start human trials—two essential steps that need to be taken before opening up the treatment to the public.
There’s also the matter of availability. While lopinavir and ritonavir are widespread, remdesivir is harder to come by, says Ho. “Remdesivir was actually developed for ebola and Gilead [the drug manufacturer] is still in clinical trials... for this drug, so it's actually not that available yet.”
Reports indicate, however, that remdesivir approval is likely to be expedited in some countries. Both Japan and the United States have approved the use of the drug for severe cases. In Europe, the European Medicines Agency has announced that it has begun a rolling review of remdesivir, and in the meantime it’s available to patients through both clinical trials and ‘compassionate use’ programmes, where patients are able to access unauthorised drugs in emergency situations.
See also: Is Tech Still Forgetting About Women?
Previously, Ho has used AI to successfully optimise drug treatments for other diseases, including tuberculosis and HIV, as well as helping to prevent liver transplant rejection in patients. All of the drug combinations pinpointed though IDentif.AI are patent-free and available to the public, Ho says. “We’re doing this because we want our results to help as many people as possible: patients, clinicians, policymakers, and beyond.”
Ho hopes his team’s findings through IDentif.AI will highlight the benefits of using AI in healthcare processes in general. Current methods of designing drug combinations are time-consuming and “fairly traditional” and, as a result, can severely limit the number of drugs, combinations and dosages tried, says Ho.
Currently, drug treatments are created by repurposing approved drugs, combining them together and looking for a dose. It’s this process, says Ho, that’s the problem. “Because when we pick two or three drugs and we go and look for the dose, we're already limiting ourselves to potentially even better options that are out there,” says Ho. Whereas with AI, you are able to search a larger pool of drugs and combinations in a far shorter time frame.
“AI is promising in that it can find actionability out of uncertainty,” says Ho. “In two weeks, out of hundreds of thousands or even a trillion possible combinations, IDentif.AI found a powerful strategy to address the coronavirus.”
“That level of speed is a major advantage when time is of the essence.”
This article was updated on May 26, 2020 to include data from a study published in the New England Journal of Medicine on May 22, 2020.