Artificial Intelligence in the Life Sciences Industry — Strategies for IP Protection

“As AI uptake and application in the life sciences continues to increase, companies, investors and patent professionals alike will face complex questions and uncertainty surrounding protection and ownership of AI-related IP. But [that] should not deter efforts to invest in, plan for, and protect valuable IP assets.”

artificial intelligence - intelligence (AI) is a rapidly growing sector of the life sciences industry, with broad-ranging applications in drug discovery, biotechnology, medical diagnosis, clinical trials, precision and personalized medicine and patient monitoring. The recent uptick of AI use in this industry is likely due to the increasing availability of “big data.” AI technologies including machine learning, deep learning, and natural language processing can be harnessed to process vast data sets to identify new drug candidates, optimize drug dosing, match patients with drug trials and diagnose diseases. Recognizing this potential, global biopharma companies have invested heavily in AI technology—the AI in life sciences market was valued at USD 1092.44 million in 2019 and is expected to reach USD 3445.60 million by 2025.

Unsurprisingly, investors in this field are keen to secure IP protection for their AI-related innovations. But unlike traditional life sciences inventions, IP protection for AI-related innovations raises fundamental questions of inventorship, patent eligibility and public disclosure, which are not readily addressed by the current U.S. patent law.

For example, last year, the United States Patent and Trademark Office (USPTO) issued a decision rejecting the idea that an AI “machine” can be an inventor. With this in mind, consider two life sciences scenarios: (1) scientists at a pharmaceutical company use AI as a tool to help screen and identify new drugs by directing AI to do specific tasks; and (2) a pharmaceutical company purchases an AI system from a third party (e.g., IBM), trains the system using a proprietary dataset, and uses the system to design a new drug with little or no human interaction. Under the first scenario, inventorship can be readily determined by identifying the scientists that discovered the new drug. However, the second scenario is more complex—who or what is/are the inventor(s) of the new drug? And who owns the related product and process IP? Although answers to these questions may not be readily apparent now, there are concrete steps that life sciences companies can, and should, take to protect their AI-related IP.

Contractually Define AI-Related IP Rights

First, life science companies utilizing AI can mitigate potential IP ownership issues by defining ownership of AI-related IP rights in employment, licensing, or purchase agreements. Such agreements should also specify who will own the data, information or results that may be generated by AI, who can use it, and how it will be used.

For example, in the second scenario above, IBM could explicitly retain all IP rights in the AI system itself but license a pharmaceutical company certain IP rights in the output of the AI system, such as new drug compounds. Further, such contracts should spell out who will be liable for the actions taken and the results obtained using AI systems now or in the future.

Consider Trade Secret Protection for AI Inventions

Second, aside from general contractual protections of IP, life sciences companies should consider what type of IP protection is appropriate for their AI-related inventions. Some AI technologies, including machine learning and deep learning, operate in part in a “black box” that may not be easily reverse-engineered. The corollary is that it can be difficult to detect, let alone prove, infringement of patents claiming machine learning technology. Under such circumstances, trade secret protection can offer advantages over traditional patent protection.

Trade secrets are not subject to patentability requirements, including inventorship, written description, enablement, and subject matter eligibility under Section 101, all of which raise unsettled questions in the AI context, as discussed below. Thus, if an AI invention is unlikely to be reverse-engineered and may be particularly vulnerable to patentability challenges, trade secret protection should be considered. Trade secrets also have the added advantage of having immediate effect (with no waiting time at the USPTO) and provide potentially indefinite duration.

However, companies choosing this route should be aware that the trade secret protection is only maintained if the owner takes “reasonable measures” to keep it secret (see, e.g., 18 U.S.C. § 1839(3)). This secrecy requirement may prove difficult for life sciences companies given the large number of individuals involved in discovery, product development, regulatory, and manufacturing activities. Thus, companies should proactively establish specific security measures—e.g., limit the disclosure of trade secret information to only those who need the access to the information and require signed NDAs that identify the trade secret information to be protected before granting access to such information. Pharmaceutical companies should also notify the Food and Drug Administration of any trade secret information in their filings to prevent public disclosure by the agency (see, e.g., 21 C.F.R. § 20.61). Further, companies must be mindful of the statute of limitations for enforcing trade secrets rights (see, e.g., 18 U.S.C. § 1836(d)) and be diligent in enforcing them.

Adopt a Playbook for Patenting AI Inventions

Third, if patent protection is more appropriate (e.g., where the requirement of secrecy for trade secret protection is hard or impossible to meet), the companies should adopt an “AI-Patent Playbook” as follows to obtain patent protection for AI-related inventions.


Although the Leahy-Smith America Invents Act (AIA) eliminated the express inventorship requirement of 35 U.S.C. § 102(f), §§ 115(a) and 116(a) still require that all inventors be identified in patent applications, and the requirement of proper inventorship is, at least arguably, still incorporated through Section 101. But the USPTO, UKIPO (and UK High Court), and EPO, have all refused to permit an AI to be identified as an inventor. Furthermore, even some life sciences companies (see, e.g., Genentech’s and Novartis’s comments on the subject) appear to differ in their views of whether and/or to what extent the current inventorship framework is workable for AI inventions (see also USPTO’s report on public comments).

Despite this lack of consensus, potential inventorship disputes can be proactively minimized by drafting patent applications in a way that requires inclusion of humans as inventors. For example, where possible, patent applications can be drafted to describe the human input in the design, training and execution of AI systems, and, similarly, patent claims can be drafted to include actions taken by humans and processes that require human intervention. Some helpful suggestions can be glimpsed from, e.g., the Federal Register Notice requesting comments on patenting artificial intelligence inventions that describes “[d]esigning the algorithm and/or weighting adaptations; structuring the data on which the algorithm runs; running the AI algorithm on the data and obtaining the results” as potential examples of “the different ways that a natural person can contribute to conception of an AI invention and be eligible to be a named inventor” and USPTO’s report on public comments on AI and IP policy (see also Genentech’s and Merck’s comments on the topic.)

Written Description and Enablement Challenges

Patenting AI-related inventions also presents challenges with regards to the written description and enablement requirements. Under 35 U.S.C. § 112, patentees must disclose to the public enough information about the invention to enable one of ordinary skill in the art to visualize it and practice the claimed invention. Satisfying such requirements can be problematic where AI has assisted in discovery of a large genus of compounds or where the AI process claimed involves complex “black box” computer programming. But life sciences patent practitioners already have significant experience in determining how to satisfy these requirements for the more unpredictable life sciences arts. Like for traditional drug patents, those drafting, e.g., applications related to the AI-assisted drug discoveries must ensure that any claimed genus is tailored or well-characterized enough so that a person of skill in the art would know how to make and use the invention without undue experimentation. It is also important to include data demonstrating activity for representative members of each drug class. For claims involving machine learning or deep learning, practitioners should include as much detail as possible in the specification about the inventions claimed (e.g., architecture of the model, training dataset and methodology, pre-processing steps for new data, etc.) and avoid “black box” schematics.

Subject Matter Eligibility

The Supreme Court’s seminal decisions in Mayo Collaborative Services v. Prometheus Laboratories, Inc., 566 U.S. 66 (2012) and Alice Corp. Pty. Ltd. v. CLS Bank International, 573 U.S. 208 (2014) set forth a two-part test for patent eligibility under 35 U.S.C. § 101, under which courts must first asks whether a claim is directed to a patent-ineligible concept, such as an abstract idea, a law of nature or a natural phenomenon. If so, courts must then analyze whether the claim adds an “inventive concept” sufficient to transform the claim into patent-eligible subject matter. The Mayo/Alice decisions have made it significantly more challenging for applicants to obtain patents on computer-implemented or software patents, and patents directed to AI algorithms (see the following white paper for commentary). Life sciences patents have also been adversely impacted, resulting in the invalidation of certain diagnostic and method of treatment claims. Disagreement amongst federal courts and within the Federal Circuit itself concerning the correct interpretation of Mayo/Alice has led to increasing pleas for Supreme Court clarification or patent legislation reform on this issue.

Despite persisting uncertainty concerning the application of Section 101, patent practitioners can take steps now that can minimize the risks of future rejections or invalidations of AI-related life science claims. One approach is to draft claims that apply the abstract idea or mental process employed by the AI in a specific or unconventional manner to achieve a practical result. For example, where AI is encompassed in a claim relating to a diagnostic method or method of treatment, patent practitioners may include an active step of administering a specific drug to the patient for the specific condition to be treated (see, e.g., Vanda Pharm. Inc. v. W.-Ward Pharm. Int’l Ltd., 887 F.3d 1117, 1134 (Fed. Cir. 2018); Endo Pharm. Inc. v. Teva Pharm. USA, Inc., 919 F.3d 1347, 1353 (Fed. Cir. 2019)). Also, since the Federal Circuit held that patent eligibility may involve underlying factual questions (see, e.g., Berkheimer v. HP Inc., 881 F.3d 1360, 1368 (Fed. Cir. 2018)), practitioners may want to consider including a description in the patent specification of how the claimed invention improves on prior art or overcomes a particular scientific problem, as it may provide an added advantage. Such description could not only increase the chances of allowance by the USPTO by highlighting the differences from the prior art to the examiner prior to any responses to office actions, but also shape a potential factual dispute, and thus reduce the chances for a successful challenge under Section 101 via a pre-trial motion in a future litigation.

Looking to the Future

As AI uptake and application in the life sciences continues to increase, companies, investors and patent professionals alike will face complex questions and uncertainty surrounding protection and ownership of AI-related IP. But current uncertainty should not deter efforts to invest in, plan for, and protect valuable IP assets. Instead, the strategies described above could be adopted to minimize the risks while still harnessing the powerful potential of AI technology.


Image Source: Deposit Photos
Image ID:204005854



Warning & Disclaimer: The pages, articles and comments on do not constitute legal advice, nor do they create any attorney-client relationship. The articles published express the personal opinion and views of the author as of the time of publication and should not be attributed to the author’s employer, clients or the sponsors of Read more.

Join the Discussion

One comment so far.

  • [Avatar for Anon]
    January 27, 2021 10:51 am

    The inventorship angle – while interesting, is legally insufficient, and it must be recognized that the ‘bottom line question’ cannot be avoided even if claim terms are augmented with elements of human inventorship.

    FULL inventorship for claims must be provided, and those attorneys prosecuting AI inventions are not ethically allowed to clench tight their eyes and allow the inventorship issue to be skirted with the game of adding pieces to a claim that may reflect human elements – especially when it is clear (as is the DABUS case) that invention was in fact non-human.

    As the discussion of that case on these boards have shown – there is a clear difference between human inventors merely using a machine, and human inventors NOT coming up with the invention. I have even laid out the mental thought experiment that eviscerates a real human person from opening a box with the invention contained therein and the fact that such a person is NOT the inventor (as well, I have also provided the racially sensitive historical analog of slave owners trying to be named “the inventor” because they happened to own the property that at that (heinous) time was both the actual inventor, and – as the argument went – was not ‘legally’ fully human enough to be the inventor under patent law. Of course, that analog is imperfect from a factual basis (both because while slaves may not have enjoyed full legal humanity, they were indeed factually human, and our laws are simply not set up for non-human rights in intellectual property — with the possible exception as I have also put forth of juristic personhood. The catch of course for the US Sovereign is that Roche v. Stanford is still good law, and that case holds that there must be an initial HUMAN inventor and that juristic persons do not count.

    I believe that the authors need to take a step back and make it clear that they are not advocating an ethically questionable smokescreen avenue.