Protecting inventions developed in silico

Our first article showed that in silico modelling is now becoming a feature of biomedical patent applications, with one direct consequence: the ability to generate new candidates, new indications or new protocols at an unprecedented speed. However, this is not the first time the biotech sector has experienced such an acceleration.

In the late 1990s, high-throughput sequencing made it possible to identify thousands of DNA fragments. Many were the subject of patent applications even before their function was known. The initial application might be rejected on the grounds of insufficient description. It could even become prior art, preventing a subsequent, more precisely targeted application from being filed to properly protect that function and the applications related to it. The same ease of generation exists today with in silico modelling. The same vigilance is required, compounded by challenges specific to artificial intelligence.

Building a solid case

Sort before accumulating. A discovery model can yield dozens of candidates. An initial screening, based on strategic relevance, the feasibility of validation and the foreseeable commercial value, allows efforts to be focused before undertaking more costly verification. A portfolio diluted by too many weak applications can end up being very expensive to maintain and may undermine the credibility of the whole. 

Calibrating the evidence. Each shortlisted candidate requires validation before resources are committed to drafting, but the extent of this varies depending on the robustness of the model, the nature of the intended effect and the scope of the proposed claims. A tried-and-tested model, applied to a specific molecular interaction, may sometimes constitute a sufficient starting point. A broad therapeutic effect, on the other hand, will most often require real-world data to support it, as in T 265/23, where the clinical simulation was based on real-world antiviral data for each component and on a test of their combination 

Document the model, not just the result. What matters is not only the candidate identified, but how it was identified. The application must, as far as possible, enable an understanding of why the result obtained is technically credible: the nature of the model, input data, relevant parameters, assumptions made, selection criteria, known limitations and, where applicable, the version of the tool used. This requirement presupposes organised traceability at an early stage: it does not replace disclosure at the time of filing but enables the drafting of a sufficiently precise application and allows the conditions under which the results were obtained to be traced several years later. In T 265/23, the model was not described in detail in the application but explicitly referred to a third-party document setting out its methodology. It was this traceable reference rather than a ‘black-box’ description, that enabled the Board to link the methodology to the disclosure. 

Provide fallback positions and a portfolio strategy. An invention arising from a model can be protected at several levels: specific compound or sequence, sub-family, molecular interaction, therapeutic indication, dosage regimen, combination of treatments, selection biomarker, or even a design method. The application must anticipate these levels from the outset, by providing sufficiently robust fallback positions in case the main claim proves too broad, if the technical effect is credible only for part of the scope, or if the prior art necessitates refocusing on a more precise contribution. However, this approach is not limited to the structure of a single application: depending on the maturity of the data, the commercial significance of the candidates and the validation timetable, it may be preferable to build a portfolio of complementary applications, each centered on a better-substantiated subject matter.

Confirm, never establish. The priority year should be used to reinforce a teaching already present in the application, not to create the invention retrospectively. Data filed after the application date may, depending on the circumstances, confirm a therapeutic use that was already credible at the time of filing or substantiate a technical effect within the framework of G 2/21. They will not replace a technical concept absent at the time of filing. The correct approach is therefore to file the application as soon as a credible foundation exists, then to use the priority year to consolidate that foundation with supplementary data, ideally targeted at the most vulnerable points.

Balancing patents and trade secrets. Not everything relating to in silico research is necessarily intended for patenting. The identified candidate, its therapeutic use or a specific combination may justify patent protection. On the other hand, certain building blocks in the pipeline, certain datasets, training parameters, selection criteria or internal know-how may sometimes be better protected by trade secrets. The strategy must therefore distinguish between what needs to be disclosed to support the claimed invention, what may be the subject of a separate application, and what must remain confidential.

Coordinate patent applications with publications and partnerships. In silico inventions often emerge in environments where information flows rapidly: scientific publications, posters, fundraising, discussions with partners, access to platforms or exchanges with service providers. Each of these stages may expose sensitive information prior to filing. An effective protection strategy therefore requires aligning the filing schedule with that of scientific, financial and commercial communications at a very early stage.

Do not overlook freedom to operate. Identifying a new candidate through modelling does not necessarily mean it can be freely exploited. Prior rights may cover the target, a family of compounds, a therapeutic indication, a formulation, a combination of treatments, a biomarker or even certain tools used in the discovery process. The filing strategy must therefore be accompanied, where commercial considerations warrant it, by an analysis of third-party rights, in order to avoid building a portfolio that, whilst solid in terms of patentability, is difficult to exploit in practice.

Across these various areas – from the selection of candidates to the assessment of evidence, from the documentation of the model to the development of a portfolio strategy, right through to the trade-offs between patents and trade secrets, the coordination of applications and the analysis of freedom to operate – a patent adviser can provide valuable support to the research team at an early stage, rather than only becoming involved once the decision to file has already been taken.

Specific considerations regarding artificial intelligence

An inventor must be human. The EPO ruled on this in the DABUS case (J 8/20 and J 9/20, December 2021). Under Article 60(1) EPC, the right to a patent belongs to the inventor or their successor in title. An artificial intelligence cannot be an inventor, nor can it confer any rights. An application that does not name a natural person as the inventor is refused.

However, it must be the right person. The designation of a human being is not always sufficient to settle the matter. In the United States, the USPTO has published guidelines specifying that a person who has merely asked an AI system to solve a problem, without making any significant contribution to the claimed invention themselves, cannot validly be designated as an inventor. Conversely, someone who has designed, built or trained the system to address a specific problem, or who has significantly reworked what it has produced, may be named as an inventor. The EPO, for its part, does not verify the reality of this contribution at the time of filing, but an incorrect designation remains a risk to bear in mind for any patent family extending to the United States. 

Check who owns what. Once the inventor(s) (as natural persons) have been identified, the question remains as to the assignment of their rights, particularly where a third-party tool or platform is involved in the process. Its contractual terms must be checked before, not after, filing: rights over the data transmitted, rights over the results produced, possible reuse of the data, exclusivity, confidentiality, retention, and the possibility of freely exploiting the inventions generated using the tool. 

The model itself can also be protected. Distinct from the invention or inventions it has helped to identify, the model itself may merit its own protection. The Enlarged Board of Appeal clarified, in G 1/19, the conditions under which a computer simulation may contribute to the technical character of an invention. Depending on the circumstances, a separate strategy may therefore be considered for the pipeline, the model’s architecture, the selection stages, or the technical criteria used to identify the shortlisted candidates. However, this protection must be devised whilst taking into account what falls within the scope of patents, trade secrets and in-house know-how. 

Confidentiality is never guaranteed. Using a third-party model to analyse sensitive data carries a risk that is often underestimated: that of inadvertent disclosure, which could destroy the novelty of one’s own invention even before the application is filed. A misunderstood sharing feature, a poorly planned data retention policy, or reuse for training purposes may be sufficient to render information accessible to the public within the meaning of Article 54(2) EPC. Before transmitting commercially sensitive data to a third-party tool, it is essential to check precisely what its contractual terms stipulate regarding the storage, reuse and rights acquired in respect of both the data transmitted and the results produced.

In summary

The lesson from genomics and the one currently emerging with AI are similar: the speed at which an invention is produced has never exempted it from the traditional requirements of patent law. Developing a robust filing strategy for in silico inventions requires combining biotech patent expertise with a detailed understanding of the models used, legal scrutiny of the contracts governing their use and an international perspective from the outset, given how practices can vary between patent offices.

This is precisely what Plasseraud IP offers: a healthcare team dedicated to the pharmaceutical and biotech sectors, a team specialising in artificial intelligence, and legal experts capable of assisting with the negotiation of contractual clauses and general terms and conditions of sale, who are particularly attuned to the specific challenges raised by the use of these tools. This cross-disciplinary approach is complemented by our capacity to provide international support, drawing both on a strong network of overseas colleagues and on in-house expertise across several key jurisdictions, notably US patent attorneys – including former USPTO examiners – as well as Chinese and Japanese practitioners. This combination enables us to anticipate, right from the filing strategy stage, the specific requirements of the main territories of protection.