AI and Public Health, Part 3: How AI Can Revolutionize Drug Discovery

This R Street series considers how artificial intelligence (AI) and machine learning (ML) systems are advancing medical capabilities and health outcomes. Previous installments explored how AI/ML systems are helping to improve the practice of medicine and address major causes of death and other ailments. This installment examines how algorithmic AI/ML capabilities aid scientists and pharmaceutical companies in developing new drugs while enabling medical professionals to better tailor drug treatments for patients.

In its latest annual Top 10 Emerging Technologies report, the World Economic Forum listed “AI for scientific discovery” as its top pick, noting how “[t]he world is on the cusp of a science discovery revolution driven by AI.” McKinsey consultants say AI/ML presents a “once-in-a-century opportunity” and predict that the technology could generate $60 billion to $110 billion a year in economic value for the pharmaceutical and medical-product industries. McKinsey identified five specific ways in which generative AI could benefit the health sciences value chain, with research and early discovery of drugs constituting the second biggest potential benefit of $15 to $28 billion of expected annual value. A 2022 Morgan Stanley study predicted that, thanks to new AI/ML capabilities, 50 novel therapies could be developed over a 10-year period, which could translate to a more than $50 billion opportunity.

What these developments mean for average Americans and their health is what is most important. While it is still early, “innovation in the field has led to massive leaps” already. Scientists have noted a surge of AI in drug discovery in recent years and predict that, due to ongoing advances, “more meaningful contributions will be made to substantially transform the field” in coming years.

Technological Revolutions Coming Together

AI/ML is helping enhance drug discovery today with recent interrelated radical breakthroughs in different fields that reinforce one another in a symbiotic fashion. In an era of rapid “combinatorial innovation,” new knowledge is constantly compounding and facilitating advances in multiple fields simultaneously. AI/ML itself rests on information and capabilities developed through innovations in many other sectors including computing, data processing systems, online networks, sensors, and more. As innovation occurs across these fields, it creates a positive feedback loop—not just for AI/ML, but for many other sectors, including life sciences.  

The author of a new book on this “superconvergence” of technologies explains how the genetics, biotech, and AI sectors are coming together as innovation begets innovation across those fields in a continually accelerating loop. These advances, he says, are “pushing our knowledge, understanding, and applications forward at an accelerating rate,” and having a particularly profound impact on human health.

From Generalized to Personalized Medicine

These developments will help us move beyond the limits of generalized medicine (i.e., treatments for the masses) and toward more personalized medicine (i.e., treatments tailored to individuals).

Developing and prescribing drugs is still a bit of a guessing game, with medical professionals hoping they can take something that worked well for many people and use it for others. But the problem with generalized medicine, of course, is that no two people are alike. Even those drugs that regularly help large numbers of people (like aspirin and acetaminophen) sometimes have serious side effects for others.

This problem arises from the fact that drug development and pharmacology—the study of how drugs affect humans and other living things—are remarkably complex and filled with considerable assumption-making. “The number of chemicals that theoretically could prove useful as drugs has been estimated at ten to the sixtieth power—a quantity greater than the number of atoms in the solar system,” one doctor noted. Therefore, generalized approaches will remain the norm until scientists find the right combinations.

AI/ML can help take some of the guesswork out of this process by lowering the search costs of finding useful new pharmaceuticals while allowing medical professionals to tailor drug treatments and better predict potential patient responses to them.

Identifying New Molecules, Proteins, and Compounds

“You do not need to talk to people in computational biology for long to understand their excitement about AI,” a Financial Times columnist recently observed. “Machines have a number of advantages over their flesh-and-blood researcher and lab assistant counterparts,” they note, which can “help advance scientific research, accelerate drug discovery and improve patient outcomes.”

AI/ML systems are already providing a better understanding of molecules, proteins, and compounds, which are foundational to chemistry and biology, and using that knowledge to revolutionize de novo drug design, which is “the generation of novel molecular entities with desired pharmacological properties from scratch.” A 2021 survey of scientific literature on recent AI/ML-driven advances in medicine concluded that “[t]he involvement of AI in the de novo design of molecules can be beneficial to the pharmaceutical sector because of its various advantages,” that can lead “to swift lead design and development.” Illustrating the influence of AI/ML in this field, the Nobel Prize in Chemistry 2024 was awarded to three researchers for computational protein design and protein structure prediction.

In this regard, one of the most important developments took place in 2022, when an ML system from Google DeepMind called AlphaFold was able to model the structure of nearly all known proteins, representing “a significant advance in biology that will accelerate drug discovery.” Before AlphaFold, identifying the shape of a single protein took scientists months or years in a laboratory, and only about 0.1 percent of known protein structures had been solved.

Thanks to this innovation, predicting the shape of almost every known protein in the world has become possible. Researchers from the Fundamental AI Research Team at Meta have a competing ML-enabled database of 617 million predicted protein structures. These advances are leading to what some researchers call a “protein design revolution” driving the “next quantum leap in the biotech industry,” which could completely transform medicine. This innovation continues with DeepMind announcing in April 2024 the latest and more powerful iteration of AlphaFold and, in September 2024, its new “AlphaProteo,” which generates novel proteins to advance drug design and disease understanding. Simultaneously, a new company called Chai Discovery released a new foundation model for molecular structure prediction called Chai-1, which enables unified prediction of proteins and small molecules to facilitate a variety of drug-discovery tasks.

AI’s societal advantages in facilitating drug design were seen most recently in the response to the COVID-19 coronavirus outbreak. In 2020, the world’s most powerful supercomputer at the U.S. Department of Energy’s Oak Ridge National Laboratory in Tennessee was used to power a computational model that conducted rapid research on more than 8,000 compounds that could help fight COVID-19. In less than two days, the model was able to narrow the list to 77 possible helpful compounds, when such research normally would have taken months. Today, the Oak Ridge supercomputer is used in partnership with the biopharmaceutical company BPGbio to conduct rapid-fire research on the company’s biobank of more than 100,000 samples to push the boundaries of personalized drug discovery. For “[t]hings that historically would have taken us six to nine months to process,” the CEO of BPGbio said, now “it’s like nine hours.”

Similarly, Pfizer scientists credit its ML “Smart Data Query” technology with allowing the clinical trial data for its COVID-19 vaccine to be reviewed a mere 22 hours after beginning the process of examining patient response data. Although this process has traditionally required Pfizer scientists to manually inspect tens of millions of data points for coding errors and other inconsistencies, one company official said the new ML-enabled process “saved us an entire month.”

While it remains to be seen whether AI can help “craft the perfect flu shot” and make a “turbocharged” flu vaccine, other recent AI/ML advances have shown promise for finding drugs to address influenza and other viruses.

Boosting Cancer Therapies

The next wave of AI/ML-enabled personalized drug creation could revolutionize cancer treatment. Because “every cancer is as unique as the individual it afflicts,” it is important that on-demand cancer vaccines “be tailored to convey each patient’s unique cancer mutations with precision,” by helping “identify patterns and correlations that traditional human-led methods might overlook.”

Fortunately, such patient-centric cancer vaccine trials are showing promise. The New York Times recently profiled a wave of startups using AI models to “remake drug discovery,” and “moving the field from a painstaking artisanal craft to more automated precision, a shift fueled by A.I. that learns and gets smarter.” Likewise, The Wall Street Journal has reported on a wave of new biomedical startups using AI “to predict the response patients will have to cancer treatments, aiming to increase the success of drugs in clinical trials and tailor therapies to individuals.” These new capabilities help medical experts sift through massive volumes of information and predict which drug treatments will work best for patients.

Conclusion

Beyond drug discovery, AI also can help drugmakers throughout the drug life cycle. For example, AI can assist industry and regulators with ex post drug safety monitoring and identifying counterfeit drugs and drug theft. Pharmaceutical companies can also use AI for market analysis, prediction, and segmentation.

These and other benefits will likely grow over time, as long as policymakers remain open to the fast-moving, unpredictable changes to come. Regulators also are trying to adapt to new AI-driven medicine. In 2023, the U.S. Food and Drug Administration (FDA) released two major reports addressing AI’s role in drug development. One of these reports explored the extensive array of agency regulations that already could cover many algorithmic systems but also noted that “[c]ontinuously learning AI systems that adapt to real-time data may challenge regulatory assessment and oversight.”

Beyond the FDA, many other government bodies and policies affect drug discovery and distribution. America’s slow and costly regulatory system will need to adapt to the AI/ML revolution because, as one leading health technology investor argued, “[i]f AI is restricted in this space, the loss to healthcare would be tragic…. We must do what we can to accelerate these desperately needed innovations.” A policy environment conducive to innovation and investment is essential to attaining that goal.

View other posts in the AI and Public Health series.