3 June 2024

Antibodies and AI: patentability considerations as de novo antibody design emerges

As Artificial intelligence (AI) begins to become more relevant to all areas of research, the technology has been applied to a variety of biotechnology areas. Antibody design has followed suit and AI has been used to improve, and more recently, generate entirely new antibodies.

Guidelines and case law relating to patent applications are constantly developing in relation to both antibodies and AI. In March 2021, the European Patent Office (EPO) Guidelines for Examination first included a section devoted to the examination of antibody-related patent claims (G-II-6), and this has been updated in 2024, as discussed here. Although this has helped to harmonise the approach of all EPO Examiners to the examination of patent applications, it does not touch upon AI generated antibodies or how they are examined specifically. As applications relating to AI generated antibodies increase in number, we may see this reflected in the Guidelines.

The concept of inventive step (or similar criteria of non-obviousness in some territories) plays a critical role in determining the patentability of antibodies. The established view of the EPO’s case law is that many aspects of the production of an antibody, including antibody generation, involve methods that form part of the skilled person’s common general knowledge. The EPO Guidelines state that inventiveness is not acknowledged solely on the basis that an antibody is structurally different from the prior-art antibodies and arriving at alternative antibodies exclusively by applying techniques known in the art is considered to be obvious to the skilled person. The fact that the structure of an antibody, i.e. its amino acid sequence, is not predictable is not a reason for considering the antibody as non-obvious (see T 605/14, section 24; T 187/04, section 11). As a result, the bar for inventive step can appear to be very high in Europe, and the vast majority of antibody cases before the EPO Boards of Appeal are decided on inventive step.

Traditionally, creating antibodies involved labour-intensive processes such as immunizing animals or screening vast numbers of molecules. Recently, AI has been harnessed both to improve and modify existing antibodies, and well-known competitors in the field include LabGenuis and AlphaFold. For example, LabGenius collaborated with Sanofi in a multi-year partnership focused on inflammation research, which has been used to build a machine-learning based model to predict protein potency and improve antibody features. Likewise, AlphaFold has focused on predicting antibody structures. Both LabGenius Ltd and DeepMind Tech Ltd have been filing patent applications relating to such methods and developments.

However, of particular interest is recent developments in de novo antibody creation. The research team at the University of Washington developed an AI tool called RFdiffusion, which allows researchers to design proteins that can strongly attach to other proteins of interest. These custom proteins differ significantly from antibodies, which recognize their targets through flexible loops that are challenging to model using AI. To overcome this limitation, the team modified RFdiffusion by fine-tuning the neural network using thousands of experimentally determined antibody structures, and designed antibodies recognizing specific regions of bacterial, viral, and cancer-related proteins. The researchers synthesized a subset of their AI-designed antibodies in the laboratory and tested their binding capabilities.

Although these antibodies generated from AI methods are new, patent applications relating to new antibodies to known targets generated using these methods would likely face a similar inventive step hurdle to those produced by more traditional methods due to the relatively little human intervention required in the antibody optimisation process.  An inventive step could, however, reside in the identification or selection of a new or refined target antigen, or in any surprising function or property of the antibody. It will be interesting to see how the case law around the patentability of such novel synthetic antibody-related proteins develops.

Please contact our specialist Antibody Team if you would like more information in this ever-progressing technical area. Please also let us know if you would like to receive our regular antibody newsletter.