15 April 20246 minute read

Generative AI, drug discovery, and US patent law

Generative AI is becoming increasingly useful in drug discovery and development, both to discover new drugs and to identify new uses for existing drugs. GenAI can be used to generate new molecular structures with desired properties tailored to target particular molecular binding sites. It can be used to virtually screen interactions between these new compounds and cellular targets. Researchers are also employing GenAI to predict a novel molecule’s pharmacokinetic profile as well as repurposing existing drugs for new uses. All of these uses allow scientists to better understand complex biological mechanisms and increase the speed at which new drugs are discovered, tested, and brought to market.

Like all new and emerging technologies, the use of GenAI in drug discovery and development leaves several open questions in US patent law. While current case law has established that an AI system cannot be an inventor for the purposes of Section 101, it has not provided much guidance as to how inventorship should be determined when GenAI has been employed.

Recently, however, the USPTO, however, issued guidance on AI and inventorship which does provide some insight as to how these issues may be treated when eventually litigated.

Inventorship

The Federal Circuit has confirmed that an inventor, as required by statute, must be a natural person – therefore, an AI system cannot be a named inventor.

In 2021, scientists and inventor Stephen Thaler challenged the USPTO’s rejection of his patent application which named an AI system he’d developed (called DABUS or Device for the Autonomous Bootstrapping of Unified Sentience) as sole inventor in the Eastern District of Virginia. The USPTO cited a series of decisions finding that only natural persons can be inventors. The Eastern District of Virgina issued its decision agreeing with the USPTO that the plain language of the statute controls and requires a human inventor, as upheld by Federal Circuit case law, and in September 2021 it declined to extend inventorship to DABUS. Thaler challenged the decision to the Federal Circuit, which upheld the decision. In an August 5, 2022 opinion, the Federal Circuit affirmed, finding that the Patent Act requires an “inventor” to be a natural person. The Supreme Court denied Thaler’s petition for a writ of certiorari on April 24, 2023.

The USPTO has been proactive on this issue, forming the AI and Emerging Technologies partnership to gather input from diverse stakeholders, including industry, academia, and independent inventors. On February 13, 2024, the USTPO published “Inventorship Guidance for AI-Assisted Inventions,” with the goal of assisting applicants in determining the correct inventors to list. The USPTO’s guidance reiterates that only natural persons can be recognized as inventors. As such at least one natural person is required to significantly contribute to each claim. AI may assist or play a significant role in the inventorship process so long as a natural person also makes a significant contribution to the invention.

The guidance instructs that the “significant contribution test” to evaluate the contributions made by natural persons in the invention creation process should be conducted on a claim-by-claim basis. This requires an analysis of each claim to determine whether a natural person significantly contributed to the inventive concept. Having a natural person merely analyze an output from an AI system is not a significant contribution of a natural person. A natural person must contribute to the conception of the invention, and reduction to practice by a natural person of an invention conceived by another is not enough to warrant listing that person as an inventor.

While courts have long analyzed joint inventorship between human beings, how courts will determine what constitutes a “significant contribution” with respect to inventions that incorporate GenAI remains to be seen. An instructive parallel, however, might be drawn between the use of GenAI and therapeutic antibody discovery. In GenAI-assisted drug discovery, the GenAI tool helps identify compounds, but human input is still required to test the compounds in vitro. Similarly, in therapeutic antibody discovery, animals or cells are infected with an antigen to induce an immune response and generate antibodies, which are then tested and optimized by research scientists. While the animals or cell lines used in therapeutic antibody discovery are what generate the potential antibodies, they are viewed as a research tool and do not test the antibodies or make the final determination as to which antibodies possess the sought after therapeutic benefits.

In this sense, existing GenAI technology is much like a research tool. While it aids researchers by identifying compounds, it does not ultimately determine which of the compounds identified possess the sought-after therapeutic benefits.

Enablement and written description under Section 112

The USPTO’s guidance is currently limited to inventorship, but notes “that AI gives rise to other questions for the patent system besides inventorship, such as subject matter eligibility, obviousness, and enablement.”

In the context of enablement and written description, patents that claim black box concepts such as “machine learning” or “neural networks” without describing the specific details of the AI model risk running into section 112 challenges. For example, courts have invalidated AI-related patents on section 112 grounds where the specification has failed to disclose how the AI algorithm performed the claimed functions.

The use of GenAI creates an additional hurdle: with much of the process of generative AI occurring in a black box that would be challenging to describe in a specification, inventions developed with the assistance of GenAI may run afoul of Section 112. could plausibly find themselves subject to challenges based on inadequate enablement and/or written description if the specification fails to adequately disclose the GenAI system in question.

Unlike a cell line well characterized in the scientific literature, each GenAI system is unique and depend on inputs and the data on which it is trained. How much of a GenAI system would need to be disclosed in the specification in order to satisfy statutory enablement and written description requirements remains an open question.

To learn more about the implications of the USPTO guidance and this evolving area of law, please contact either of the authors or your usual DLA Piper relationship attorney.

Print