Artificial Intelligence (AI) is making significant strides in a field once considered a holy grail of biological sciences—protein structure prediction. The significance of these breakthroughs was highlighted by the Nobel Prize in Chemistry, awarded to the creators of Google’s AlphaFold, a generative AI (GenAI) model capable of predicting protein structures with unprecedented accuracy. Over the past year, funding and collaboration within the biopharma sector have surged with the aim of leveraging the power of AI platforms to accelerate discovery and design of complex biologics, such as antibodies. Current deep learning (DL) models can reliably predict the structures of proteins and research is now focusing on de novo antibody design. These antibodies are fully developed and selected by AI against specific targets or epitopes while also addressing critical factors like affinity, cross-reactivity, and developability. The emergence of de novo antibody design intersects with two rapidly evolving areas of patent law—AI and antibodies— raising important questions about how to secure effective patent protection given the current challenges in patenting these biologics.
Are de novo designed antibodies patentable?
De novo designed antibodies can be patented, but there are important considerations. While the antibodies themselves may be patentable, the AI methods used to create them rely on mathematical models and algorithms. In most major jurisdictions, these are excluded from patentability, as they fall under the broad category of abstract mathematical concepts. Therefore, to patent algorithms/mathematical models it is necessary to demonstrate the technical/ tangible effect of the algorithm to overcome this barrier to patentability. It is crucial to include the technical advantages and practical applications of the invention in the patent application to ensure it is not deemed unpatentable.
Who is the inventor of a de novo designed antibody?
Courts in numerous countries have ruled that the terms ‘inventor’ and ‘inventorship’ are reserved for human inventors, meaning that in most major jurisdictions, an AI system cannot be recognized as an inventor[1]. However, uncertainty remains regarding the criteria for determining inventorship in AI-assisted inventions.
The United States Patent and Trademark Office’s (USPTO) ‘Inventorship Guidance for AI-assisted Inventions’ (‘USPTO Inventorship Guidelines’)[2] offer some insight, defining an inventor as someone who has made a ‘significant contribution’ to the invention. The guidance largely relies on the Pannu factors[3] to determine inventorship in AI-generated inventions. The key takeaway is that a human inventor must provide additional inventive input to meet the criteria for inventorship. Such contributions might include, for example, modifying an AI-designed antibody in vitro to optimize its antigen-binding properties, or training the AI system to identify antibodies specific to certain targets. Similarly, the European Patent Office (EPO) also does not recognize non-human inventors, and while the EPO does not provide similar criteria for determining inventorship in AI-assisted inventions, the focus is more on AI-assisted inventions achieving a specific technical purpose and demonstrating a technical effect to achieve said purpose[4].
How can de novo designed antibodies address inventive step?
Inventive step is likely to present the biggest challenge moving forward as identifying antibodies for specific targets becomes increasingly easier. Patent Offices are unlikely to consider a designed antibody inventive solely because it was created by an AI system. For instance, the USPTO Inventorship Guidelines require that inventors demonstrate a significant contribution to the invention that goes beyond the mere prompting of an AI system to generate results. Therefore, the skill and ingenuity of the inventor in manipulating the AI system or optimizing the antibody experimentally to meet certain requirements or improving antibody properties, are among the key factors to consider when considering the inventive step of a de novo designed antibody.
How much is enough? The sufficiency conundrum with antibody patents
Most jurisdictions require that a patent disclosure demonstrate that the inventor was in possession of the invention at the time of filing the application and enable a person skilled in the art i.e. a person reasonably proficient in the scientific area, to carry out the invention. Thus, the disclosure must have sufficient details of the invention to meet these requirements. It is also important to keep in mind that the patentability requirements of a de novo designed antibody and the AI-system that built the antibody are distinct, therefore, the data required in support of each aspect must be considered individually.
When considering patenting antibodies, an antibody may be defined, broadly, in a patent claim by its (i) amino acid/nucleic acid sequence, (ii) the target/antigen/epitope it binds to, and (iii) by its properties such as binding affinity to the target. Patent claims that describe the latter two are called functional claims as they describe the functional features of the antibodies. It is a broader claiming strategy as any antibody that binds to the target and/or with said functional features will fall within the scope of such a claim. To meet the sufficiency requirement, patent specifications would normally describe and exemplify the properties of a few antibodies displaying the claimed functional features. Over the last decade, however, patent offices in many jurisdictions have become averse to such broader claiming trends and rejected claims for lack of sufficient support from the specification. This has steadily led to the replacement of functional claiming with sequence-based claiming, that are narrower in scope.
In the United States, for instance, the Supreme Court’s decision in Amgen v. Sanofi[5] (and post-Amgen decisions) has set the precedent for disfavouring functional claiming of antibodies under grounds of lack of enablement. The Court's decision was essentially based on the assumption that identifying the antibodies encompassed in Amgen’s claims would not be routine for a person of ordinary skill in the art and place undue burden on the skilled person due to the extensive trial-and-error required. For context, Amgen's specification described 26 exemplary antibodies that bound to the target of interest (PCSK9 protein) but the Supreme Court found that even this was insufficient guidance to enable the skilled person to carry out the claimed invention. A similar trend is also seen in India, where functional claims are unlikely to be granted by the Patent Office.
The situation in Europe is slightly different where the EPO is not opposed to functional claiming[6]. In general, the EPO takes the view that obtaining/identifying antibodies is routine and does not place a burden on a person skilled in the art. However, the EPO requires that the skilled person should be able to unambiguously identify the antibody(s) that is within the scope of the claims and be able to replicate/manufacture the exact antibody. This often requires that, along with the functional features, the method(s) and/or the detailed experimental parameters required to identify the functional characteristics be included in the claims.
As it stands today, the preferred route for patenting antibodies is by claiming the specific CDR sequences of the antibody as the likelihood of an application being objected under grounds of lack of sufficiency is reduced. However, with the evolution in AI technology the sufficiency bar set for antibodies by patent offices will most likely need to be reconsidered. While previously identifying an antibody with the desired characteristics would involve trial and error and screening several candidates, using AI systems such as Google's AlphaFold3, this process is greatly simplified and the argument for ‘undue burden on the skilled person’ would no longer be applicable. In the case of de novo antibodies, functional features such as the epitope of the target antigen would be sufficient for an AI system to identify suitable antibody candidates within minutes in the future. Therefore, it remains to be seen whether patent offices will be more accepting of functional claiming and broader antibody claims in the era of AI-generated antibodies.
The AI methods used to design antibodies would be subject to the same data requirements that are currently in place for existing AI inventions and computer-implemented inventions. In general, in order to fulfil sufficiency requirements, an AI method must be described in as much detail as required for a person skilled in the art to replicate the method. In particular, care must be taken to provide details of the steps that are crucial for the invention as claimed- particularly the inventive step. For instance, while describing the training data for an algorithm may not be necessary in all cases, if the method in the patent claim is dependent upon the training data to achieve the desired output, then this aspect must be disclosed as well.
Overall, the ease with which antibodies can be designed in the future could allow for broader claiming strategies than exist today as the sufficiency hurdle would be easier to overcome. This would require, however, that the AI methods and essential design parameters that are used to obtain such antibodies are described in as much detail as possible in the specification. Also, considering that AI inventorship criteria require demonstration of a significant contribution from the human inventor, it may also be necessary to supplement the AI program data with practical experimental examples showing the superior properties of candidate de novo designed antibodies.
Conclusions
The implications of de novo antibody design are yet to be fully realized by the biopharma sector and the legal sector. While significant progress is underway in the development of this technology, the law is still catching up. Certain aspects of the law such as inventorship and patentability would expect to see little change, but others such as inventive step and sufficiency would need to be reassessed if innovation in this sector is to be encouraged. Overall, the emergence of de novo designed antibodies marks an exciting era for both the pharmaceutical industry and the legal landscape.
[The authors are Consultant and Executive Partner, respectively, in IPR practice at Lakshmikumaran & Sridharan Attorneys]
[1] Refer to the DABUS case in various jurisdictions, e.g. Thaler v Vidal 43 F.4th 1207 (2022) in United States, or J 0008/20 (Designation of inventor/DABUS) 21-12-2021 at EPO.
[2] Inventorship guidance for AI-assisted inventions, effective 13 February 2024 (Refer Example 2).
[3] Pannu Factors: Inventor must (i) contribute in some significant manner to the conception or reduction to practice of the invention, (ii) make a contribution to the claimed invention that is not insignificant in quality, when that contribution is measured against the dimension of the full invention, and (iii) do more than merely explain to the real inventors well-known concepts and/or the current state of the art.
[4] Refer to Part G, Chapter II, 3.3.1 of the EPO Guidelines for Examination (2025).
[5]Amgen Inc. v. Sanofi, 598 U.S. 594 (2023)
[6] Refer to Part G, Chapter II, 6.1 and 6.2 of the EPO Guidelines for Examination (2025).
