wechat_bg

2020-05-04 07:24:55 Voices

This article is part of an ongoing series in which experts will analyze the role of science and technology in epidemic response and control in China and around the world.

On Feb. 4, China’s Ministry of Industry and Information Technology announced perhaps the first government initiative in history to apply artificial intelligence to an epidemic. The proposal called on the country to, among other things, “fully explore the application scenarios of AI in the diagnosis, treatment, and containment of COVID-19.”

From viral analysis and AI-aided diagnostics to scanning crowds for signs of fever, China has utilized artificial intelligence on a large scale over the past four months. But AI’s most promising application yet may be in the field of pharmaceutical research, where it’s fueling studies into new drugs for COVID-19 while also helping researchers check if any existing medications could be repurposed for the fight.

My company has been using computational methods to understand the virus and search for potential cures since Jan. 20, the day the new coronavirus’ genetic sequence was released to the public. Thus far, the industry has focused most of its efforts on checking existing anti-viral medications for their potential effectiveness against COVID-19. Compared with starting drug development from scratch, finding new uses for existing drugs offers a potential shortcut to effective treatment, in part because the selected drugs have already undergone rounds of rigorous testing and clinical trials to establish their safety and side effects.

Using our AI-powered drug research and development platform, our team screened for possible mutations in the virus. We then screened more than 2,900 Food and Drug Administration-approved drug molecules and over 10,000 traditional Chinese medicine molecules for candidates that could potentially inhibit three of the protein subunits the virus uses to transcribe itself on host cells and replicate. Using high-accuracy computational chemistry algorithms, on Feb. 2 we narrowed our list down to the 38 most promising candidates. We are currently collaborating with pharmaceutical companies and research institutions to test their efficacy in experiments.

The repurposing of existing drugs for use in treating COVID-19 is not without limitations. Because they were not specifically designed to target the new coronavirus, they are likely to be less efficient, meaning they might require higher dosages to be effective and consequently cause stronger side effects. Some of the most initially promising repurposed drug candidates were an AIDS cocktail consisting of lopinavir and ritonavir, Ebola treatment candidate remdesivir, and the anti-malarial drug chloroquine. All of them have suffered setbacks in recent clinical trials, however, as researchers seek to ascertain their efficacy and safety in COVID-19 patients.

The powerful computing capabilities offered by AI can help us significantly broaden our search for potential cures while speeding up the process.

There are seven strains of coronavirus known to cause mild to potentially life-threatening diseases in humans, including COVID-19, severe acute respiratory syndrome, and Middle East respiratory syndrome. Unfortunately, previous attempts to develop drugs and vaccines for SARS and MERS fizzled out, because the epidemics subsided before drug development was complete, making it impossible to recruit enough patients to complete the necessary clinical trials.

If we don’t want to keep repeating this cycle, we need to design and develop broad-spectrum anti-coronavirus drugs capable of fending off any future mutations. The process of developing a new drug is long. Researchers first need to understand the biological mechanisms that lead to an infection or disease, then they must identify and design drug molecules that can effectively bind to proteins in the virus or in the body, preventing the virus from replicating itself. After that, through a series of experiments of increasing complexity — first on a biochemical and cellular level, then on animals — researchers carry out tests into these molecules’ antiviral activity and safety. Finally, promising drugs are tested in a clinical setting to verify their overall efficacy in humans.

Chemists posit that there at least 10^60 potential molecules with drug-like properties that could one day be tapped for pharmaceutical research. Currently, we have access to up to 10^23 of these for drug discovery and research. Trying to find new molecules with certain properties and effects on a given target in such a vast field is a gargantuan task, but the powerful computing capabilities offered by AI can help us significantly broaden our search for potential cures while speeding up the process.

The advantage of applying AI in drug discovery is not limited to its computing power. AI algorithms also have a strong ability to learn iteratively. We can teach an AI the “language” of biology and chemistry, then use it to discover laws and patterns that are beyond human powers of perception or comprehension. An AI model can generate a massive amount of molecules or amino acid sequences, then design new sets of molecules based on experimental or computational feedback. Relying on the interaction between AI and human experts, we can select a small group of the most promising candidates and subject them to further experimental testing.

In our own drug screening efforts, we’ve chosen to focus on the coronavirus’ spike protein — which plays a crucial role in infecting human cells — as well as a few other key protein subunits that are widely considered attractive drug targets. We hope to identify molecules that are effective in treating COVID-19: ones whose efficacy, toxicity, and key physicochemical properties make them more likely to succeed. We’re also exploring biomolecules, which are formed by sequences of amino acids.

Pharmaceutical research and development is a complex, long-term, and high-risk undertaking that demands sophisticated technology and considerable investment. Traditional pharmaceutical R&D takes, on average, more than 10 years to create a drug from start to finish and can cost upward of $1 billion. For every 5,000 to 10,000 molecules that enter the R&D pipeline, only one will become a marketable drug.

Much of that time is spent in clinical trials, but by combining AI with other cutting-edge computer technologies and experimental techniques, we hope to shorten the early-stage drug discovery process from three to four years to just one year. The global research community is in a race to find effective therapies for COVID-19, but thanks to the development and application of AI and other algorithm-driven technologies, we are hoping to see more good news regarding the advancement of drug and vaccine development over the next three to six months.

It’s possible many of these projects will fail or not be completed in time, but the insights we accumulate and the new technologies we develop will empower pharmaceutical innovation and hasten our response in the future.

Translator: Lewis Wright; editors: Wu Haiyun and Kilian O’Donnell; portrait artist: Wang Zhenhao.

(Header image: Wang Zhenhao for Sixth Tone)