In early May 2024, Google DeepMind launched AlphaFold 3, an advanced AI model capable of predicting the structures and interactions of biological molecules. The upgrade generates 3D models for proteins, DNA, and other molecular entities, aiding scientists in making groundbreaking new drug discoveries.
In recent years, leading companies such as Pfizer Inc., Eli Lilly, Novartis, and Merck & Co. have already forged partnerships with AI firms. Analysts project that advancements in AI and machine learning could lead to the development of approximately 50 novel therapies over the next decade, potentially generating over $50 billion in pharmaceutical opportunities.
Regulatory bodies are endeavoring to keep pace with these rapid advancements in AI technology. During this transitional period, drug sponsors incorporating AI into new drug applications and patents should stay vigilant regarding potential infringement issues and what might be deemed prior art.
Artificial Intelligence in New Drug Applications
In 2016, FDA’s Center for Drug Evaluation and Research (CDER) received just one investigational new drug application (IND) incorporating artificial intelligence (AI) elements. By 2021, this figure had surged to 128 submissions containing AI components. To date, CDER has processed over 300 such submissions, the majority pertaining to oncology, with notable representation in gastroenterology, psychiatry, and neurology. The AI technologies cover an array of therapeutic areas and span various stages of drug development, from initial discovery to post-market safety monitoring.
The primary objective behind integrating AI into drug discovery is to expedite the development of transformative treatments while reducing associated costs. Enhanced accuracy and accessibility in obtaining three-dimensional models of complexes enables researchers to more effectively conceptualize antibodies, nanobody biologics, or small-molecule drugs that interact with specific therapeutic targets.
DeepMind Technologies Limited—a British-American AI research subsidiary under Google—has significantly advanced the previous AlphaFold 2 with AlphaFold 3. This latest iteration extends capabilities beyond protein structure prediction to include modeling DNA, RNA, and other molecules such as ligands.
In roughly one-third of cases analyzed thus far, structures predicted by AlphaFold 3 have been instrumental in significantly accelerating project timelines—potentially cutting years off traditional experimental methods required for obtaining new structures and providing a considerable strategic advantage in pharmaceutical research and development.
Patentability of AI-Generated Drug Innovations
Still, questions continue to emerge regarding the novelty and patentability of AI-generated innovations in drug discovery. An invention that is already publicly known is considered prior art. Since 2021, AlphaFold's predictions have been openly available through a database containing over 200 million protein structures and have been cited in others’ work thousands of times.
While AI can accelerate the discovery process exponentially by generating numerous potential molecular structures, it also can create an expansive repository of prior art. This could hinder subsequent innovation before anyone has determined the utility or relevance of the prior art.
Should AI-generated prior art prevent pharmaceutical companies from securing patents for active pharmaceutical ingredients (APIs), companies may become reluctant to invest in early-stage R&D efforts that could bring new treatments.
In February 2024, the U.S. Patent and Trademark Office (USPTO) issued guidance clarifying that while AI can assist inventors, inventions solely generated by AI are not eligible for patent protection.
According to the Guidance, inventions facilitated by AI can be patented if significant human contribution is deemed sufficient for inventorship. The USPTO has yet to offer specific direction on managing AI-generated material as potential prior art.
Drug developers should proactively address possible obstacles by publishing their research and varying descriptions of their inventions. Introducing an additional element or limitation may prove crucial in navigating disputes involving potential AI-generated prior art.