AI Helps Pharma Find New Drugs But Imperils Lucrative Patents – Bloomberg Law

6 minutes, 10 seconds Read

Artificial intelligence has emerged as a double-edged sword for pharmaceutical companies, helping to develop new cures while also creating obstacles to obtaining drug patents potentially worth billions.

A vast trove of AI creations would make it increasingly difficult for companies to convince patent offices around the world their inventions are novel, minimizing their incentives to invest the massive sums necessary to find new treatments, attorneys say.

Pharmaceutical clients are already facing what they suspect is AI-generated prior art—the term for evidence showing that an invention is already known and therefore ineligible for a new patent—during the patent application process, attorneys said. Litigation opponents could easily cite similar material to invalidate patents they’re accused of infringing.

“AI can put discovery into hyperspeed and spin out various potential molecules,” said Robin Feldman, a professor at the University of California College of the Law, San Francisco. But, she said, it has “the potential to create a huge pool of prior art blocking downstream innovation, long before anyone’s figured out what to do with that prior art, or whether it’s even useful.”

If AI-generated prior art precludes pharmaceutical companies from obtaining a patent for an active pharmaceutical ingredient—the component in the medicine that causes the desired effect—they’ll be less likely to put up the money for even early-stage R&D that could lead to new treatments.

“No one’s going to develop a drug if they can’t patent a drug,” said Nicholson Price, a health law and intellectual property professor at the University of Michigan. “That’s the general rule now.”

Losing those investments could mean the public doesn’t even see what new drugs they’re potentially missing, McDonnell Boehnen Hulbert & Berghoff LLP Partner Joshua R. Rich said.

“Pharmaceutical companies are not beloved by many people in the world,” acknowledged Rich, who has represented pharma companies in patent disputes. It’s possible activists or others “may very well leverage the tremendous power of large language models to undercut the value of pharmaceuticals.”

The US Patent and Trademark Office issued guidance in February making clear AI can be used to assist inventors, but wholly AI-generated inventions aren’t eligible to be patented. The office hasn’t yet offered guidance on how it will handle AI-generated material that could be prior art, though it pledged to study the issue.

PTO Director Kathi Vidal has discussed the “big problem” of AI-generated prior art at least twice in the last month, including at Nvidia Corp.‘s March 20 AI conference, where she specifically mentioned AI-generated molecules.

“If you just put it out there in the universe and have AI come up with every possible solution, but you don’t do the testing—you don’t do the real work, you don’t do the investment to actually get that into people’s arms to solve for cancer, to solve for new diseases—that’s not helpful,” Vidal said.

She told Nvidia general counsel Iain Cunningham that the PTO wants to know whether AI-generated content should be considered prior art. She previously said the office will issue a request for comments on the issue, first mentioning an inquiry as far back as September. The PTO declined to comment on a timeline.

R&D Impact

The pharmaceutical industry pumps billions of dollars each year into the R&D process with the goal of discovering and delivering new drugs, therapies, and medical devices to the market.

The leading lobbying group for drug manufacturers, the Pharmaceutical Research and Manufacturers of America, reported its member companies over the last decade have doubled their annual investment in search for new treatments, spending nearly $101 billion in 2022 alone.

Recognizing how AI can speed up the drug-discovery process by designing drugs, proteins, and molecules, companies including Pfizer Inc., Sanofi SA, and Merck & Co. have signed deals with AI companies in the last few years. Morgan Stanley estimates AI and machine learning could lead to an additional 50 novel therapies over 10 years, which could translate to an opportunity exceeding $50 billion.

“A lot of pharma companies are already using AI for drug discovery, and if they’re not doing it, they’re planning to do it,” said Benjamin Hsing, a partner at Venable LLP.

But the speed at which AI can generate prior art references may simultaneously hinder some pharmaceutical innovations and cut promising candidates from the drug pipeline.

Even if AI-generated creations don’t make it into patent applications, they can be used as “defensive publications” to impede the patenting of discoveries where companies have dedicated resources to development. Defensive publications are details distributed into the public domain to stop others obtaining a patent on the same invention.

They are often cited in various stages of the patenting process, including by patent examiners evaluating applications and by opponents seeking to invalidate patents at the Patent Trial and Appeal Board. Websites like and TDCommons already serve as platforms for defensive publishing.

AI could be leveraged to generate more defensive publications, Sterne, Kessler, Goldstein & Fox PLLC Director David H. Holman said, including every variation it can make of claims in existing patents.

“You could imagine using some of these tools to identify potential therapeutic compounds, and making it harder for others to patent those things,” Wolf, Greenfield & Sacks PC shareholder Dan Rudoy said.

But drugmakers aren’t helpless against the threat. Patent applications based on AI-generated content can be easy to spot because they can lack traditional research methods and provide experiment data that’s purely “in silico”—performed by a computer, instead of in a test tube or in animals—Rich said.

“When you don’t see that human involvement in the applications, especially when you see only prophetic examples, not actual research that has been done on the compounds, that’s a pretty big giveaway that they haven’t actually done the research after developing it,” he said.

AI-generated prior art can also be nonsensical, Holman said. The machine may just be cranking out and rearranging words.

“Every now and then it’s going to hit and these words are going to make a sort of coherent disclosure,” he added.

Looking Ahead

While lawyers wait for the PTO to act, they’ve started to fashion workarounds for clients facing AI-generated obstacles.

Because AI-generated references are usually just text, Holman said, clients can try to anticipate and overcome them by including experiments, working examples, and data. He’s also considered different ways to describe inventions because he said adding one extra element or limitation could be the missing piece in potential prior art generated by AI.

“It can give you that extra hook that says, ‘Well, this AI doesn’t disclose all the elements of what we’re claiming here,’” he said.

In the litigation context, there could be evidentiary hurdles to introducing AI-generated prior art. Parties will need to prove when a document was published and whether it was publicly accessible, details that could be unknown or challenged by opponents, Holman said.

AI-generated publications “can be nonsense,” he added, but the pharmaceutical industry is monitoring the technology’s development.

“Every week we’re learning more, it’s getting better, and smarter, and faster, and growing,” he said. In a matter of time there will “probably be more and more of these references that have more sort of coherent types of disclosures. So it’s definitely on our radar.”

This post was originally published on this site

Similar Posts