Drug discovery involving artificial intelligence (AI) tools has quickly occupied significant territory in the pharmaceutical industry. One study found that the number of startup drug candidate pipelines employing AI is roughly equivalent to 50% of the preclinical programs of big pharmaceutical firms. See, Jayatunga et al., “AI in Small-Molecule Drug Discovery: a Coming Wave?” Nature Reviews, March 2022. The prevalence of AI has generally led to significantly reduced drug discovery timelines. Current research data indicates that AI-driven discovery pipelines on average reach the preclinical phase within four years, compared to the conventional expectation of five to six years. Id.
Despite the values brought to the business, the rapid implementation of AI might have created unintended effects in law that could severely impact a pharmaceutical company’s right to the drug. The IP rights in AI-driven drugs, like those in drugs discovered using conventional methods, will mostly take the form of patent exclusivity before the generic market is open to competitors. Yet, AI use in drug discovery is still early enough, that if AI “discovers” the drug, the state of law has not yet been established to address whether the pharmaceutical company will enjoy a similar exclusivity. Two recent case decisions, despite not being related to drug discovery, are examples signifying that there can be circumstances where a pharmaceutical company may not be entitled to the same exclusivity. In a copyright registration case, the U.S. Copyright Office has denied the registration of an artwork named “SURYAST” that was generated by AI, finding insufficient human authorship in the creative work. On the patent front, in Thaler v. Vidal, 43 F.4th 1207 (Fed. Cir. 2022), the Federal Circuit held that the term “inventor” in the Patent Act refers to a natural person, and, thus, AI cannot be an inventor. The logical extension of the holding of Thaler is that, if AI is deemed the sole inventor of a drug, the drug will be ineligible for patenting under 35 U.S.C. §101, which states, “whoever invents … may obtain a patent ….”
This article explores the benefits and risks of AI-driven drug discovery from the legal perspective. Since the law governing IP rights in AI-driven drug discovery is still in its infant state, any future legal development is likely to have significant implications in many areas. For example, new court or legislative outcomes can affect a pharmaceutical company’s considerations in integrating AI tools into its discovery pipelines, M&A diligence of a big pharma acquisition of a pharmaceutical startup, a licensing deal between a pharmaceutical company and an AI company that provides AI tools, and day-to-day practice of scientists’ R&D journaling and publication.
The General Overview Governing AI Involvement in Drug Discovery
The above recent IP law decisions appear to allude to a general concept that IP rights will not be awarded if AI completely or significantly replaces human ingenuity in the creative process. To elaborate, if AI were merely used as an acceleration tool in drug discovery cycles, it should not be treated differently from other computer or laboratory tools that have existed for decades. Hence, the researchers will still be the rightful inventors. Yet, if the researchers cross an undefined threshold by allowing AI to replace the human role of inventing, IP rights may be threatened.
However, the actual application of this general concept will likely be difficult and uncertain until more case law has developed. Both the drug development process and the legal framework of inventorship determination for a drug are highly complex. For example, drug discovery is a highly iterative process. Unless the process is fully replaced by a series of AI models and automated robotic agents, there will likely always be involvement of both humans and AI. To understand to what extent the human involvement becomes so insignificant that the pharmaceutical company will lose the patent right, we will first discuss the typical process of drug discovery and the basic legal framework of inventorship determination in this context.
Typical Drug Discovery Process
Drug discovery is typically a complex and iterative process that often involves certain main stages, including target discovery, lead identification, and lead expansion and optimization. These stages, combined, often span five or six years to generate a viable drug candidate and involve a large number of researchers, including chemists, biologists, biostatisticians, and engineers.
Target discovery often involves identifying a disease’s mechanism. The target identified usually is a protein, an enzyme, or a gene with which a future drug will aim to interact to alter biological pathways governing the disease’s proliferation, thereby treating the disease. The drug candidate that will interact with the target can be a small molecule organic compound, an antibody, a polypeptide, etc.
Lead identification is a stage where screening is performed on a multitude of compounds to ascertain their efficacy at interacting with the target. However, due to the sheer amount of structural diversity leading to potential drug candidates and the inherently unpredictable nature of molecular interaction, researchers with a thorough understanding of the disease mechanism almost certainly still need to test a large number of candidates.
Lead expansion and optimization typically involve extensive iterative cycles of chemical structure optimization, biological testing, and study of structure-activity relationships, aiming to refine a lead compound into a prospective drug candidate.
Various AI tools can be integrated into different stages of drug discovery. In general, AI tools are often developed to shorten the amount of screening and iterations, but can also be deployed for any purpose in the discovery process.
What Constitutes Inventing in Drug Discovery
The legal framework of inventorship determination is best explained based on the Inventorship Guidance for AI-Assisted Inventions (AI Guidance) issued by the U.S. Patent and Trademark Office (USPTO) published in February 2024. Under the AI Guidance, the USPTO has made clear that, “while AI-assisted inventions are not categorically unpatentable, the inventorship analysis should focus on human contributions, as patents function to incentivize and reward human ingenuity.” The key principle set forth in the AI Guidance is that “patent protection may be sought for inventions for which a natural person provided a significant contribution to the invention.”
However, this significant contribution must be made to the “conception” of the invention. The law has established a dichotomy of “conception” and “reduction to practice” in determining inventorship. The law is clear that conception, not reduction to practice, is the touchstone of inventorship. Conception is often referred to as a mental act that involves the formation in the mind of the inventor of a definite and permanent idea of the complete and operative invention. Reduction to practice is often referred to as the process of bringing the mental conception to a tangible form to demonstrate its operability. The AI Guidance emphasizes that “a human perform[ing] a significant contribution to reduction to practice of an invention conceived by another is not enough to constitute inventorship.”
This dichotomy distinction is not straightforward in drug discovery because drug invention rarely involves an instantaneous lightbulb mental moment that a candidate is the right drug. The iterative process involving scientists’ strategic thinking and lab experiments makes it difficult to determine when conception is completed. Now, in the context of AI, for a patent right to be granted, there must be human involvement and significant contribution before the conception is completed solely by AI. A conception is completed when the inventing entity has formed the idea in a sufficiently final form and has a reasonable belief that the invention is operative. This sounds similar to a situation where a machine learning model (inventive entity) outputs a compound formula in digital form (sufficiently final form) with a predicted affinity score representing the affinity for a target enzyme (appreciation that the compound is operative in interacting with the target enzyme). As such, under a superficial application of inventorship law, it seems that companies are at high risk of losing patent rights because, in some situations, conception could be deemed completed at the moment AI digitally outputs a chemical structure or formula. In such situations, humans have no opportunity to contribute to the conception.
However, case law developed in the small molecule discovery space signifies that the use of an AI candidate screener does not always automatically doom a patent right. The Federal Circuit has held that the conception of a chemical compound requires both the idea of the compound’s structure and the possession of an operative method of making it. Amgen, Inc. v. Chugai Pharm. Co., 927 F.2d 1200, 1206 (Fed. Cir. 1991). For inventions related to compounds whose syntheses are unpredictable, courts have held that the conception of a compound requires the inventor to recognize how the compound can be isolated from the rest of the physical environment. This has led to a doctrine called the simultaneous conception and reduction to practice that is unique to inventing a chemical compound. Pursuant to the doctrine, in certain situations where synthesizing a compound is unpredictable, an inventor may only be able to establish conception after pointing to a reduction to practice through a successful experiment. In other words, conception is not completed unless the researcher successfully demonstrates that the compound can be isolated in the wet lab. If this simultaneous doctrine applies, it mitigates the risk of losing patent rights because it reduces the chance of conception being deemed completed at the AI output. AI only outputs a digital representation of the compound, not a physical sample of the compound. The completion of the conception of a compound that is unpredictable to synthesize still requires a researcher to demonstrate that the digital representation can be produced and isolated in physical form.
Given this doctrine, the completion of the conception of a chemical compound seems to hinge on when the inventing entity has determined how or knows how to synthesize the compound. For some compounds whose method of making is well understood, the conception may have been completed at the moment AI generates the digital representation of the compound. For other compounds that are more complex and unpredictable to prepare, the conception may not be completed until the researchers can demonstrate that they have successfully isolated the compound in the wet lab. In other words, when conception is completed will depend on the difficulty in converting a digital representation of the compound outputted by AI into a physical sample of compound. As such, whether AI will become the sole inventor of a compound (and thus, the pharmaceutical company may lose some of the patent rights) may also depend on the ease with which a compound can be synthesized.
The Risk-Reward Dilemma
Logically, this leads to an observation that the risk level of losing patent rights with AI involvement is different for different classes of drugs. For example, antibodies and polypeptides may become the highest risk class for a company to lose patent rights if AI outputs the sequence and there is no subsequent human alteration of the sequence. The reason for this is that there are well-established synthetic methods for preparing antibodies and polypeptides of almost practical sequence. In fact, oftentimes scientists send the digital sequences to a third-party vendor to make research samples of the antibody or polypeptide to avoid the need to synthesize the antibody or polypeptide themselves. Hence, at the moment that AI generates an amino acid sequence, it is almost certain that the sequence can be synthesized in the lab. In this type of setting, a court in the future may hold that conception of the sequence is completed by AI alone.
This presents a great dilemma to pharmaceutical companies in deciding whether to adopt an AI tool in discovering antibody/polypeptide drugs. The dilemma is present because AI is quite suited for accelerating the identification of an amino acid sequence, yet this class of drug has the highest risk of losing patent rights. A polypeptide is a linear sequence of amino acids that can be synthesized easily based on the sequence because of its linear nature. Yet, the number of variations of a polypeptide is virtually infinite because each amino acid site has 20 potential substitutes (when considering only canonical natural amino acids), so even only 10 amino acid sites have 2010 possibilities. As such, AI is suited in this space to screen and virtually simulate a large number of variations of a polypeptide. However, as discussed, conception may have been immediately completed the moment the AI outputs a polypeptide sequence. A pharmaceutical company could face the tough decision of deciding whether to adopt a powerful AI acceleration screening tool whose outputs may confer weaker IP rights than results that are generated by a non-AI laboratory method.
The risk for a small molecule drug is lower for using AI tools in screening. A small organic molecule is typically three-dimensional in its overall structure (instead of a linear sequence), and the manner of synthesis is typically considered by courts as unpredictable and can vary molecule by molecule. Hence, a synthetic organic chemist will likely need to be involved in determining how the digital formula outputted by the AI can be made into a physical form, thereby giving the organic chemist a contribution to conception under the doctrine of simultaneous conception and reduction to practice. From the IP perspective, employing another AI model to replace the organic chemist’s role in determining how the organic molecule should be synthesized would likely increase the IP risk.
Joint Contribution Will Likely Be the Predominating Fact Pattern
The previous section addresses the risk in the scenario when the AI tool luckily (or unluckily) outputs a formula or sequence that is precisely the final drug candidate. The more common situations would involve some human optimization of the AI output to alter the structure of the compound before a final drug candidate is selected, such as in the lead expansion and optimization stage. Hence, the more likely fact pattern will be that the AI and a human jointly contribute to the invention of the drug.
The joint contribution scenario can occur in various situations. For example, first, AI may generate a few species of a class of molecules that can be used as a drug while a human conceives of other species in the class based on the optimization of AI-generated species. Second, a human researcher may input a broad class of molecules into an AI model to screen for a few promising species in the broad class. Third, within a polypeptide, AI may identify amino acid substitutes for a few sites, while humans may identify substitutes for other sites. Fourth, AI may contribute to an ingredient of a pharmaceutical composition while humans may invent the overall composition. Fifth, AI may provide an upstream contribution to an invention while humans may contribute to certain downstream ideas, such as determining the dosage of a molecule identified by the AI.
From the patent law perspective, the above examples can be generalized into two ways of claiming an invention. The first two examples may belong to a situation where a patent includes some species claims and a genus claim that cover both AI-generated species and human-generated species. The latter three examples may be situations where a patent claim includes an element A conceived of by AI and an element B conceived of by a human.
From the perspective of analyzing the risk of losing patent rights, the law tends to be more settled regarding the first way of claiming (genus-species). It is rather well-established that the conception of a species within a genus may constitute a conception of the genus, but a conception of a genus is not a conception of a particular species in the genus. Hence, the law will likely remain the same, such that a pharmaceutical company can claim a genus that includes a human-invented species optimized from an AI output because the human-invented species alone can support the genus claim. However, the pharmaceutical company may still run the risk that it may not be able to broadly cover a claim directed particularly towards the AI-invented species in a given genus if the AI-invented species happens to be the best-performing drug candidate.
The law seems to be more unsettled with respect to the second way of claiming (a claim with elements A & B). The USPTO, in the AI Guidance, currently takes the position that although AI cannot be named as a co-inventor, “the inability to list an AI system, used to create an invention, as a joint inventor does not render the invention unpatentable due to improper inventorship.” In other words, the USPTO appears to imply that a human inventor inventing element B but not element A in a claim can support the patentability of the entire claim. Put differently, under this USPTO’s position, a company will not lose its patent rights even if AI conceives of an indispensable element A in the invention as long as a human significantly contributes to element B. However, this USPTO position has not been tested in the courts, and the agency’s stance may eventually be rejected in litigation. In fact, the USPTO emphasizes that the AI Guidance is “iterative” and “does not have the force and effect of law.” Since pharmaceutical patents are highly valuable, even a small chance of losing patent rights warrants substantial consideration from a pharmaceutical company.
AI-Pharma Startup Considerations
AI startups in the drug discovery sector should consider the above analysis to position themselves for success. This patent issue can have implications on fundamental issues such as the exit strategy of AI-pharma startups.
For example, if an AI startup company opts for positioning itself as an acquisition target for big pharmaceutical companies by developing a drug candidate in certain circumstances in-house using its powerful AI tools, some loss of exclusivity on the drug candidate could amount to a deal killer. As such, the startup company should engage IP counsel early in its drug discovery process to ensure that the process is properly structured with sufficient human contribution to the drug candidate. Startup co-founders tend to be pioneers who like to fundamentally change an existing process. While a fully automated AI drug discovery process may be groundbreaking from the business and marketing standpoint, startup companies should balance innovation in R&D methods with the risk of a material adverse event in IP.
The risk-reward balance may also play a role in publications. Often publication of papers serves an important purpose in marketing, in sales, and in rewarding scientists. But a company should manage the content of its publication through the lens of IP rights. While touting the versatility of an AI model or a largely automated drug discovery process is a great way to generate interest from investors, partners, and customers, companies should take care to avoid the authors inadvertently overemphasizing, for marketing purposes, the role of AI in such a way as to minimize or suppress human inventive activity in the drug development process. In addition to implicating USPTO duty of candor considerations, such explicit or implicit diminution of human contributions may promote unfavorable IP outcomes for the invented drug. It is not inconceivable that a startup in the future will proudly write an article on its groundbreaking fully automated drug discovery process that generates a successful FDA-approved drug without knowing the ramifications of such an article on its IP rights.
Big Pharmaceutical Company Considerations
From the business and research standpoint for big pharmaceutical companies, the benefits and possibilities bought by AI tools would likely significantly outweigh the IP risks described in this article. Hence, big pharmaceutical companies should understand the risk profile involved in using AI instead of completely eliminating AI in the drug discovery process. The risks would arise in various legal aspects, including in the adoption and licensing of AI tools, in the M&A acquisition of a startup with a drug candidate, and in day-to-day drug discovery operations.
In adopting AI tools, a pharmaceutical company should understand that AI tools have odd and unique risk profiles that have not been previously understood. An individualized AI pipeline that is specific to each discovery project generally would increase human contribution and reduce IP risk. Yet, given the complexity of the drug discovery process and the law in this area, there is no one-size-fits-all solution. A pharmaceutical company should engage IP counsel early to ensure AI tools are properly used.
In licensing an AI tool from a third-party AI company, a pharmaceutical company should understand that there is a risk of inadvertently partially losing IP rights to that AI company. The AI Guidance from the USPTO indicates that there are potential scenarios where the trainer of the AI can become a co-inventor of the drug, such as in a situation where a data scientist’s fine-tuning of a model is deemed a significant contribution to the invention. If the parties intend to be in a pure SaaS subscription relationship, the pharmaceutical company should include in the license a term in which the AI company contractually assigns all potential rights in the drug to the pharmaceutical company and agrees to cooperate with the pharmaceutical company if someone from the AI company is deemed an inventor. Such clauses are not always standard in typical SaaS license agreements.
In acquiring a pharmaceutical startup company that has a drug candidate, a big pharmaceutical company should consider adding an investigation of the involvement of AI as a standard diligence item. Currently, standard due diligence practice usually does not involve discussions about how AI was involved in company processes, review of academic papers published by the startup for discussion of AI use, or study of lab notebooks that documents scientists’ contributions versus possible AI contributions. As AI becomes more prevalent in the drug discovery process, thorough diligence review of IP rights, including involvement of AI, will likely be a part of any M&A. Having a deep understanding of the legal nuances at the intersection of life sciences and AI will be essential.
In-house counsel in pharmaceutical companies may have tough decisions to make in determining whether and how to recommend further documentation of AI involvement in research in addition to scientists’ existing lab notebook practices. On one hand, a more detailed record of experimentation can bring benefits such as in final drug candidate selection for executives to make more informed decisions and greater ability for counsel to assess legal risks. Such documentation may also be necessitated by the obligation to disclose material facts to the USPTO in a patent application process if AI’s involvement becomes a material fact that affects the patentability of an invention.
On the other hand, depending on the extent of AI involvement, a typical AI tool can generate a large number of results in each run. What exactly a scientist should document is not straightforward to decide. Subjecting every AI output to some legal scrutiny will likely be disruptive to the research process, but relying on traditional lab notebook practice without any deliberate effort in documenting AI’s involvement could bring uncertainty to the IP risk profile. What manner of documentation is preferable will depend on the precise situation of a research team in using the AI tools.
International Risk Profiles
The above analysis is largely based on the state of US law. The risk profiles could be drastically different in foreign countries. For example, in Thaler, the patent applicant noted that he was able receive a granted patent in South Africa listing AI as the inventor. The development of law related to AI inventorship in different countries can be policy driven, depending on how a country would incentivize AI development.
Conclusion
The implications on IP rights from the use of AI in drug discovery will likely become an important area of law development that will bring concrete and significant economic impact on various players in the pharmaceutical industry. Understanding AI’s role in the IP issues early in the drug discovery process will become critical in ensuring a company’s success. It will be vital to develop a holistic exclusivity strategy that maximizes a company’s IP rights in light of AI involvement but also to understand Al tool adoption and due diligence in M&A.
*****
Fredrick Tsang is an Associate in the Silicon Valley office of Fenwick & West LLP. Fredrick’s practice focuses on strategic intellectual property counseling and patent portfolio development related to artificial intelligence and the intersection between AI and life science. He can be reached at ftsang@fenwick.com.
Antonia Sequeira is a Partner in the Silcon Valley office of the firm. Antonia’s practice is uniquely focused on life sciences companies with multidisciplinary technologies at the intersection of biotechnology, device and/or high tech, including AI. She can be reached at asequeira@fenwick.com.
Carl Morales, Ph.D., is a Partner in Fenwick’s New York office. Carl applies his scientific and legal training toward acquiring, protecting and managing patent rights for clients in the biotechnology, chemistry, pharmaceutical and medical device industries. He can be reached at cmorales@fenwick.com.
Disclaimer: Reprinted with permission from the May 2024 issue of The Intellectual Property Strategist. © 2024 ALM Media Properties, LLC. Further duplication without permission is prohibited. All rights reserved.
Also published by Law.com.