As the COVID-19 pandemic continues to infect people across the world, a technological application already familiar to many in the biotech field is lending a key supporting role in the fight to treat and stop it: artificial intelligence (AI).
AI is currently being used by many companies to identify and screen existing drugs that could be repurposed to treat COVID-19, aid clinical trials, sift through trial data, and scour through patient electronic medical records (EMRs). The power of AI in COVID-19 is that it is being used to generate actionable information—some of which would be impossible without AI—much more quickly than before.
A simple definition of AI is the ability of a computer to rapidly think and learn. AI utilizes machine learning to analyze large amounts of data. It can also model predictions, screen virtually and develop insights that can be used to advance R&D and make patient medical assessments.
AI hasn’t always been embraced by the biotech industry, says Ronald Dorenbos, PhD, MSc, director and founder of consulting firm BioFrontline and a former head of innovation and scouting at Takeda. “A lot of the conservatism that was there in the industry is now kind of disappearing,” Dorenbos says. “Even some of these people that were against using [AI and machine learning], they now see the advantages.”
AI in repurposing drugs
One promising AI application is repurposing existing drugs that could be effective in treating COVID-19, in particular probing molecular structures. Russ Altman, MD, PhD, co-director of the Stanford Institute for Human-Centered AI, says that AI methods can be used to probe the three-dimensional pocket of critical proteins in the viral infection pathway.
“Increasingly in the last ten years, because we have so much data…many of the 3D-structure predictions and the drug-interaction prediction algorithms wind up being based on machine-learning technologies,” Altman says. “We use some AI methods to predict the 3D structure of a human protein that is used by [SARS-CoV-2] to infect and then we use another set of machine learning methods to try to find existing drugs that would fit into that pocket that we modeled in order to generate hypotheses.”
In another tack, Altman notes that using AI to examine EMRs can also help in repurposing drugs for COVID-19. The question there is: “Do I see patterns in the medications the patients are on, the pre-existing diseases they have or other elements of their EMRs that predict when they do poorly and when they do well?”
Altman explains that if a molecular prediction points to a small-molecule drug, and medical record databases show that patients taking that drug for another condition are not showing up in the ICU as would be expected, that’s interesting. “If we see that those same drugs are underrepresented in COVID-19 severe patients, that might suggest there’s been [an experiment] already done on these drugs indicating that they might be useful. These would give you a lot of confidence you’re on the right track.” (Altman held an information session on Quora in late April to answer questions about AI and COVID-19).
370 to one
One company that has used AI to look for existing drugs to test as COVID-19 treatments is UK-based BenevolentAI. Through its research, the company identified the rheumatoid arthritis (RA) drug Olumiant (baricitinib) from Eli Lilly as a promising candidate. Last month, BenevolentAI and Eli Lilly announced that the National Institute of Allergy and Infectious Diseases has started a randomized-controlled trial of baricitinib to examine whether the drug can stop COVID-19 from infecting the lungs and reduce inflammatory damage.
Peter Richardson, PhD, BenevolentAI’s vice president of pharmacology, told GEN that the company first started work on repurposed drugs for COVID-19 in January following the outbreak in China. The endeavor marked two firsts for the company: the first time its AI technology was not used to discover a novel drug and the first time it was used to find an antiviral treatment.
Early on, BenevolentAI used universal language model technology to scan through medical and scientific literature that might provide clues to how biological processes influence coronavirus. From there, Richardson identified a cluster of genes that he recognized mediated a mechanism by which viruses can enter a cell. “And embedded in them were two genes that just leapt out at me, which are kinases.”
Richardson’s team came up with about 370 known compounds that would interact with the kinases, of which 30-40 are approved by regulatory authorities. The number was further trimmed based on affinity, whittling the number down to half-a-dozen drugs. Two cancer drugs were excluded, Richardson says, based on the advice of Justin Stebbing, PhD, MA, a company consultant and oncology professor at Imperial College London, who advised that patients wouldn’t tolerate such drugs.
Richardson’s goal was to identify drugs that were anti-inflammatory and had the potential for viral inhibition. This narrowed the candidates to a pair of molecules, of which baricitnib was viewed as the most promising, based on affinity and its clearance from the body.
Richardson explains: “The advantage of drugs cleared through the kidney is that they seldom interact with other drugs, so you don’t get drug-drug interactions. We’ve got something here that controls viral entry. We can mix it with directly acting antivirals like remdesivir.”
After Richardson’s group published a letter on baricitinib in February in The Lancet, followed by more data in The Lancet Infectious Diseases, Eli Lilly got involved.
“The hypothesis of an anti-viral mechanism was inspired by BenevolentAI’s predictions, and triggered the global scientific community to collaborate on lab experiments to validate the hypothesis,” Anabela Cardoso, MD, MBA, global development lead for immunology at Eli Lilly, tells GEN.
One hypothesis is that baricitinib inhibits the host-cell protein that plays a role in viral reproduction, Cardoso says. “Baricitinib may be capable of reducing or interrupting the passage of COVID-19 into target cells and inhibiting the JAK1- and JAK2-mediated cytokine release that are increased and contribute to the complications of this viral infection.”
Germany’s Innoplexus is another player in the AI drug repurposing space. The company has identified the combination of the much publicized anti-malarial drugs hydroxychloroquine (or chloroquine) with remdesivir or the RA drug Actemra (toclizumab) as potentially effective against COVID-19. Remdesivir can help to reduce viral load while hydroxychloroquine or chloroquine will support the body’s immune system, according to Innoplexus founder and CEO, Gunjan Bhardwaj, PhD, MBA.
Bhardwaj says that the company’s AI platform continuously takes in terabytes of published and unpublished data, organizing it into knowledge graphs spanning more than 31 million biomedical terms and concepts. “Relationships and behavior of all available compounds, whether approved or under investigation, around identified pathways and mechanisms leads to generation of hypothesis around repurposing or combination therapies,” Bhardwaj says.
AI’s involvement in designing new drugs
Other companies are also using AI in COVID-19 drug discovery, but to identify new drug candidates. French AI startup Iktos and SRI International, a non-profit research institute in Menlo Park, California, recently announced a collaboration agreement to accelerate development of novel antivirals.
Nathan Collins, PhD, SRI’s chief strategy officer for biosciences, says that the goal of the partnership is ultimately to reduce the time to develop a target molecule to do preclinical and clinical testing from 3-5 years to a matter of months.
SRI has designed several automated systems for rapid reaction design to design molecules using AI to identify the best synthetic route. “We take that synthetic method and apply it to a very rapid reaction screening system that uses inkjet printing to screen the chemical methods that we’ve predicted on our AI systems to say what will actually work vs. what we will need to do more work to make actually happen,” Collins tells GEN.
SRI then uses what has been learned from the inkjet printing method “to make the molecules in multistep at whatever scale we want, from micrograms to grams of material.”
SRI had a head-start on COVID-19: about two years ago, it started research into targeting endonucleases in influenza. This produced compound and data sets they could share with Iktos, which could build an AI model to predict new molecules with more refined parameters for potency, selectivity and PK properties.
Other applications for AI
India-based Qure.ai is taking what it dubs an “AI Powered Pandemic Response” to assist those taking care of patients with COVID-19. Its qXR tools can screen chest X-rays for signs of the disease and monitor its pression, while the qScout platform—which is accessible on a smartphone app—can be used for contact registration and tracing, a chatbot for virtual triage of COVID-19 contacts and can map COVID-19 hotspots, which can aide in a public health response.
By analyzing the location, size, and type of abnormalities in a chest X-ray observed by an algorithm, qXR can classify the risk for a patient of developing COVID-19, explains Pooja Rao, PhD, co-founder and head of R&D at Qure.ai. This can help physicians measure responses to medications and optimize treatment plans at different stages of the disease.
COVID-19 is also having an impact on the way clinical trials are being conducted, not least of which is the ability of patients to go to test sites for monitoring and to receive medication. AI is also playing a role here as its ability to analyze broad amounts of data can produce insights and inform decisions in ways that were not possible in the past, notes Fareed Melhem, MBA, who heads Acorn AI Labs by Medidata, a Dassault Systèmes company.
“Overall, COVID-19 will increase the rate of digital transformation,” Melhem tells GEN. “The virtualization of trials and direct patient interaction will become much more commonplace; sponsors and site staff will have access to even more metrics and data remotely than available previously.”
Medidata has also demonstrated the use of synthetic control arms (SCA), which can create control arms for studies based on historical control data statistically matched to baseline characteristics of an experimental arm. Since many studies have issues enrolling enough patients and many patients in the control arm of a trial often drop out, an SCA can alleviate this issue. Medidata presented a case study involving non-small cell lung cancer to demonstrate the usability of SCAs at last year’s ASCO conference.
The future of AI
The rapid adoption of AI will provide evidence for what works—and what doesn’t—for health applications. “It’s ‘trial by fire’ because you’re deploying AI for a new disease in new environments—healthcare systems that may not have been previous users of AI—at a pace that tests not only the accuracy of the AI software, but also its ability to handle rapidly increasing volumes of medical data,” says Rao. “If it passes this test, AI could well be here to stay.”
But Stanford’s Altman cautions that while AI is a contributor to advances in treating COVID-19, it’s not the star of the show, merely a piece of the technology puzzle. “When we get through this, we will look back and see AI did act as a team member with the other technologies and contribute to a solution coming faster than we would have otherwise,” Altman says. “This could be a moment where we saw how to use AI in a constructive way to help with these big medical and biomedical challenges.”
