Artificial Intelligence and Public Policy to Accelerate Antibiotic Discovery

 

Recent advances in artificial intelligence (AI) present a transformative opportunity to address some of the world’s most pressing challenges, particularly in the realm of health care. While every new technology entails some level of risk, generative AI also has the potential to save millions of lives. In particular, it could help address one of the most urgent global health threats in recent history—the crisis of antimicrobial resistance (AMR).

The steady growth in drug-resistant “superbugs” over the last few decades points to a near future in which antibiotics no longer defend us from deadly pathogens. In such a world, modern medicine as we know it would cease to function. Forestalling this future demands an aggressive effort to invent, manufacture, and distribute new and better antibiotics. That task has proven difficult for a host of reasons, many of them economic. With AI, however, superbugs may have met their match.

By vastly accelerating the antibiotic discovery process, generative AI—together with sound public policy—could play a part in solving the life-and-death struggle against AMR. This effort, however, requires collaboration across sectors: Social ventures driving innovation, academic researchers uncovering new scientific insights, and government support to bring these breakthroughs to market. It’s an all-hands-on-deck challenge where coordinated action can turn AI’s potential into real-world solutions, ensuring we stay ahead of this growing health crisis.

The End of Modern Medicine

It’s difficult to overstate the scale and severity of the AMR threat. A new report from The Lancet estimates that between now and 2050, antibiotic resistance will kill close to 40 million
people worldwide and play a role in nearly 170 million
more global deaths.

A world without effective antibiotics is one in which all kinds of medical procedures—from hip replacements to caesarian sections, cataract surgeries to heart transplants—are no longer safe. Even minor injuries and illnesses, like skinned knees and ear infections, can develop into life-threatening infections when they don’t respond to antibiotics.

In many ways, this future is already upon us. Each year, more than 2.8 million
drug-resistant infections occur in the United States. And according to a report released earlier this year by the Centers for Disease Control and Prevention (CDC), rates of hospital-acquired infections from seven drug-resistant pathogens—among them Carbapenem-resistant Enterobacterales
and multidrug-resistant Pseudomonas aeruginosa—increased by 20 percent
during the COVID-19 pandemic.

Without a new generation of medicines to fight drug-resistant infections, existing antimicrobials will continue to grow less effective. And yet, for years, promising efforts to invent and market new ones have fallen prey to a uniquely broken economic ecosystem.

Bridging the Second ‘Valley of Death’

For scientists and researchers in drug development, there’s a saying known as the “valley of death.” It refers to when a scientific discovery is successful in the lab but can’t advance to human clinical trials, be it from a lack of funding or scientific success. All drug candidates that make it to FDA approval, including antibiotics, have to bridge the valley of death to get there.

And there are difficulties every step of the way. Bringing a new medicine to market can take over a decade and often costs more than a billion dollars. For many companies, the investment of time and money works out well when they secure FDA approval for medicines with large markets, such as cancer drugs or new treatments for heart disease. But antibiotics are different.

Widespread use of antibiotics increases the risk that bacteria will become drug-resistant superbugs. As such, these medicines need to be used judiciously. Additionally, even newly approved brands of antibiotics have pressure to be priced similar to older, low-cost treatments, given the highly generic nature of the market. Given these factors, the sales volume of a specific antibiotic is lower than average and expected sales revenues are meager. As a result, the chance that a company will recoup development costs is usually near zero. So, it’s immensely difficult for start-ups to raise enough capital to take novel antibiotics from the laboratory to the hospital pharmacy.

This is one of the reasons why new antibiotics that earn FDA approval have trouble making it to patients. Their developers often tumble into this second valley of death post-approval—a financial hurdle that’s unique to the antibiotic market. Of small companies that have earned FDA
approval
for a new antibiotic since 2017, all but one
have filed for bankruptcy, gone out of business, or been sold to another firm.

What’s more, none of these new treatments are in a novel class of antibiotics. This means they kill bacteria the same way existing antibiotics do, rather than a new method that could better outsmart resistant strains.

One way to address some of the unique challenges of antimicrobial research and development would be to simplify and accelerate the process of discovering new antibiotics. And that’s a job for which AI is well-suited.

Many life-science companies—including Pfizer, AstraZeneca, and Janssen—already deploy traditional predictive AI to streamline drug development. Common uses include identifying potential chemical entities for closer study, finding subjects for clinical trials, sorting through trial data, and even drafting documents and reports during a drug’s FDA approval process.

Generative AI allows for a huge leap forward that could open a new world of possibilities for accelerating antibiotic development. Specifically, it can discover potential antimicrobial agents earlier and faster than was previously feasible.

Unlike traditional AI, which draws on large quantities of data to make recommendations, generative AI can produce new compound structures, or the molecular building blocks needed to develop antibiotics. This creative capacity helps companies like mine, which are committed to bringing novel antibiotics to market.

With the help of our academic partners, we’re employing generative AI models to simplify the otherwise labor-intensive task of antibiotic discovery. In fact, my company recently received funding to optimize our generative AI platform and advance over a dozen novel AI-designed antibiotics against high-priority pathogens.

Now, a bit about how the generative AI platform operates. First, our researchers introduce thousands of different chemicals to a specific pathogen of interest, gathering data about which ones fight the infection, and which don’t. The training data are then incorporated into the generative AI platform developed by our collaborators at the Collins Lab at the Massachusetts Institute of Technology.

The platform is then put to work, analyzing the millions of chemical characteristics in the dataset to identify patterns and chemical relationships consistent among molecules with proven antibacterial activity. Then, using those patterns and characteristics, the generative AI platform comes up with new structures in silico (virtually simulated on the computer) for a potential antibiotic against the target pathogen. With the narrowed-down subset of compounds, our team runs physical tests on the candidates to see which are worth pursuing.

This technique saves money as well as time. Typically, it costs up to $10 million and takes about 4.5 years to usher a new medicine from drug discovery to the pre-investigation stage. Based on our early work, we estimate that generative AI could cut that timeframe down to just 2.5 years, while slashing the cost by two-thirds.

Moving forward, we are further enhancing the platform’s design capabilities by gathering training data on specific drug attributes so that we can not only design novel chemical structures predicted to have strong antibacterial activity, but also “filter out” those structures with high predicted levels of toxicity or specific drug absorption concerns. These training datasets will be shared in an open-access database for researchers to learn from and use in their own AI-driven antibiotic discoveries.

In this way, generative AI could soon knock down some of the barriers to bringing new antimicrobials to patients. But even these impressive gains won’t be enough to bridge the second valley of death—at least not completely.

AI Is No Substitute for Public Policy

To fully tackle the crisis of AMR, policy makers need to find new ways to help developers bridge the second valley of death post-approval and sustain the market for antibiotic development. The US Congress is considering a promising proposal that would do just that.

The PASTEUR Act would create a subscription-style system whereby the government would contract with biotech companies for a novel antibiotic. The government would provide predictable annual payments to the company in return for access to any amount of the new medicine for use in federal health-care programs, be it small or large.

By decoupling an antibiotic’s financial success from sales volume, the reform would revolutionize the economic incentives that currently govern the market for these life-saving medicines. In so doing, it could spark a renaissance in antibiotic research and development, giving the medical community an enormous leg up in the race to defeat superbugs.

The United Kingdom is already deploying a similar model. Their initiative began in 2019 as a pilot program, which identified two candidate antibiotics for government subscriptions. In May of this year, the UK government gave the go-ahead to turn the experiment into a permanent program—making it the first official antibiotic subscription system anywhere. Japan and Canada are considering similar pilot programs as well.

The combination of social ventures pioneering progress in areas such as generative AI, academic researchers advancing new discoveries, and targeted public policy could give humanity a real advantage in the fight against AMR. But it’s still early days. More companies need to embrace this new tool for developing antimicrobials, and more nations need to support these efforts through creative incentive schemes like the PASTEUR Act.

Looking ahead, generative AI could revolutionize medicine by helping us develop new antibiotics. It also has enormous potential in recommending treatment options, identifying risks, speeding up vaccine development, and enhancing disease detection, like cancer, with greater speed and accuracy. In the near term, with these combined efforts across the continuum of research and development, we can outpace resistance and defeat the AMR crisis—but the time to act is now.

Read more stories by Akhila Kosaraju.