For better or worse, AI in the pharmaceutical industry has reached fever pitch—the point of no return. Ever vigilant, the Food and Drug Administration (FDA) has stepped in to regulate the growing use of AI in drug discovery and development, vowing to “promote innovation and protect patient safety.” But while the agency is mulling the policy, companies like Immunai have already paid up front. The biotech company recently made a $18 million deal with AstraZeneca, which aims to use the immune system's AI-powered immune system model to improve clinical trials. Startups like Insilico Medicine and Recursion Pharmaceuticals are touting AI as their secret weapon in the hunt for new drugs, though skeptics whisper that their claims sound like laundry AI.
“There is a range of opinions on whether AI will solve the problems in drug development on its own, compared to whether it will work and not happen at all. I think I live somewhere in between,” Raviv Pryluk, founder and CEO of PhaseV, told the Observer. Pryluk, who previously served as SVP of Operations and analytics at Immunai, started out on his own in 2023 launching a company focused on using AI and machine learning to streamline clinical trials. “It can bring enormous value, but only if done carefully,” he warns.
The FDA says that since 1995, it has received more than 300 shipments of drugs and biological products containing AI components. Yet the complexity—and promise—of AI has grown exponentially, said Dave Latshaw II, former head of AI drug development at Johnson & Johnson. “It's a little bit different these days, especially with the scale and quality of data we have access to,” Latshaw tells the Observer.
In fact, AI is already reshaping clinical trials, from matching patients and subjects to targeting specific populations. But there's a catch: while narrowing patient types may yield more accurate results, it also limits the drug's commercial appeal. “They want to find the exact type of person that their drug will be useful for but, at the same time, doing so limits the scope of the total population that will be useful and limits the overall commercial impact,” Pryluk said.
Latshaw, who left Johnson & Johnson in 2020 to found BioPhy, sees a growing tension between technological advances and commercial priorities. He launched BioPhy to address inefficiencies he couldn't touch within the walls of a J&J office. For him, the real promise of AI is in refining the entire drug discovery pipeline, including optimizing clinical trials.
Mergers and acquisitions also present an exciting opportunity. Pharmaceutical giants often rely on acquisitions to bolster their portfolios, and AI can sift through lines of preclinical data and clinical data to identify the most promising drugs. “Machine learning and AI can help run over preclinical and clinical data and all the available literature to make better decisions about whether to get this drug versus that drug,” Pryluk noted.
AI can also help differentiate clinical studies, a long-standing challenge in the industry. Since 1993, the National Institutes of Health (NIH) has mandated the inclusion of women in research, but gender gaps—especially in preclinical animal studies—remain a stubborn problem. Researchers can combine studies using advanced methods such as causal machine learning to ensure that findings apply across populations. “What if we could use sophisticated and granular methods like causal machine learning, for example, and ask causal questions about whether the results we saw in sets of female patients would tell us something, and then we could expand the population. ?” Pryluk asked. “I think this kind of approach can bring more women, more people from minorities, to participate in clinical trials.”
Pryluk believes that precision medicine—treating each patient as unique—will be the next seismic shift.
Yes, there are still many doubts. All applications of AI in pharmacy must survive a risk-benefit analysis. Data sharing, cybersecurity, unreadable algorithms and biases pose endless challenges. For example, genetic data—the lifeblood of many AI models—raises significant privacy concerns. Just look at the ongoing debates on social media sites like 23andMe.
Steven Aviv, CTO of Pentavere, sees data trust as very important. Pentavere's DARWEN AI sifts through unstructured healthcare data to identify patients for specific treatments. “Pharmaceutical customers tell us that trust in the data of these applications is important,” Aviv tells the Observer. That's why Pentavere relies on Databricks' Data Intelligence Platform, he says. A secure framework is designed to handle large volumes of data without compromising compliance.
Latshaw, too, emphasizes transparency in AI solutions. Interpretation is important, he says, noting that researchers risk misinterpreting AI results without it. “There's always an incentive to publish the next best method for something, but sometimes you have to ask if the question you're talking about in the research is actually the right question,” warns Latshaw. “Just because you can predict the structure of a protein doesn't mean you can accurately predict what its function will be and how that affects the disease.”
Pryluk is not specific about the stakes. “We are not in the game. We are talking about patients. People are dying and it has to work,” he said.
Despite the inevitable obstacles, Latshaw is bullish on AI's long-term impact. “There will be failures in the space as always happens when people push the envelope,” he said. “But there will be an incredible amount of learning from that, and the product will have high productivity in finding new molecules.”
As the FDA debates the rule, it invites input from industry stakeholders. Latshaw reminds us that the agency's goal is to protect the patient, not to prevent industry. “If they see that the industry wants to use new technology, it is their duty to ensure that it is used in a way that does not affect that goal,” he said. “The strength of that regulation will largely depend on how much impact the technology will have on the end recipient.”
In Latshaw's opinion, the benefit could be revolutionary: in the next 5-10 years, he expects an increase in effective drug candidates to reach clinical trials. “What that means for the companies is that they will be able to bring more medicines to the market, but also get more medicines to the patients who need them,” he said.
