New pharmaceutical molecules, new clinical protocols: what impact does in silico have on patentability in biotechnology?

When modeling replaces the laboratory in the biomedical field

An “in silico” invention is an invention obtained by implementing a computational, statistical, mathematical or artificial-intelligence-based model, rather than by laboratory experimentation. In the biomedical field, this may concern the identification of new molecules or new therapeutic targets in drug discovery, the repositioning of a drug for a new indication, or the definition of new clinical protocols, for example new dosage regimens or new treatment combinations.

These methods are driven by new players in biomedical research, who no longer work only at the bench, but also on models. AI-assisted drug discovery platforms now identify both new therapeutic targets and the molecules capable of acting on them. Their first candidates are beginning to progress through clinical trials. Computational tools for predicting protein structures now reportedly make it possible to model molecule-target interactions in a matter of hours. Physiologically based pharmacokinetic simulations and quantitative systems pharmacology simulations could, in turn, predict the behavior of a drug in the body, including absorption, distribution and metabolism, without the need for a dedicated clinical study.

This development necessarily has an impact on patent cases. Applications filed in the biomedical field increasingly incorporate data from these models at an early stage, sometimes instead of traditional experimental data. A central question therefore arises for applicants: can an invention essentially established by calculation or modeling be validly protected before the European Patent Office (EPO)?

Case law specifically addressing the patentability of in silico inventions in the biomedical field remains scarce. Applications that genuinely rely on these tools are still recent, and several years will be needed before a significant number of cases reach the Boards of Appeal. A 2025 decision nevertheless provides an initial indication.

Sufficiency of disclosure: the general framework

Article 83 EPC requires the application to disclose the invention in a manner sufficiently clear and complete for it to be carried out by the skilled person at the filing date, without undue burden. A fundamental deficiency present at filing cannot be remedied later: it is possible, during the proceedings, to provide additional data supporting a teaching already present in the application, but not to create a teaching that was initially absent.

The level of requirement depends on the type of claim.

For a structurally defined product, it is generally sufficient to provide the skilled person with the means to produce it.

For a functionally defined product, the requirement is stricter: the description must provide either concrete means for obtaining it, such as a protocol or screening method, or sufficient guidance, typically a structure-activity relationship or a known biological mechanism, so that the functional scope claimed genuinely corresponds to the technical contribution disclosed. The absence of an example in the application is not, in itself, fatal. If the preparation of the product falls within well-established routine methods, such as immunization, combinatorial screening or standard cloning, the absence of experimental data is not enough, by itself, to establish insufficiency. The real test is undue burden: the Boards distinguish routine screening, even if lengthy, from undirected trial-and-error research, which is characteristic of insufficiency and is well identified in the case law relating to “reach-through” claims.

For a therapeutic use, obtaining the therapeutic effect is itself treated as a functional feature of the claim: it must therefore be credible. The application must disclose the suitability of the product for the claimed therapeutic use. A mere verbal statement is not enough: the application must provide concrete information, typically experimental tests showing that the claimed compound has a direct effect on a biological mechanism specifically involved in the disease concerned.

The patent system takes account of the difficulties inherent in drug development: it does not require absolute proof that the compound will one day be approved as a medicinal product in order for it to be claimed as such. The therapeutic effect may therefore be established by any type of data, provided that it clearly and unambiguously reflects the claimed effect. Clinical data are therefore not systematically required: an effect demonstrated in vitro may suffice if it directly and unambiguously reflects the claimed therapeutic application, or if there is an already established and accepted relationship between the observed physiological activity and the disease concerned.

It is this credibility standard that the Board clarified in T 609/02: if the description provides only a vague indication of a possible medical use for a compound that has not yet been identified, more detailed evidence produced later cannot remedy the fundamental insufficiency of the disclosure. Once this minimum evidence is present in the application, post-published evidence may be taken into account, but only to support what is already there, never to establish sufficiency on its own.

T 265/23: sufficiency of disclosure tested against in silico evidence

It is in this 2025 decision that the principles laid down in T 609/02 find a concrete application in relation to an invention essentially established by modeling.

T 265/23 (Combination of two antivirals for treating Hepatitis - AbbVie) concerned a patent claiming the combination of glecaprevir and pibrentasvir against hepatitis C, over sixteen weeks, without interferon or ribavirin. The opponent challenged sufficiency of disclosure, arguing that no clinical data were provided and that the examples did not, in its view, demonstrate any real effect of the combination on the mechanism involved in the infection.

The application did not develop a novel methodology. It relied on conventional in vitro antiviral data for each of the two compounds taken individually, as well as on an actual test of their combination on viral replicon cells, the synergistic effect of which was statistically confirmed using a long-published model, the Prichard and Shipman model. It also applied, to the two claimed compounds, a mathematical clinical simulation model whose methodology was incorporated, by cross-reference, from a third-party document cited in the application, namely a US patent application. According to the application, this model made it possible to predict sustained virologic response rates, the clinical criterion for cure of hepatitis C, on the basis of in vitro data and scientifically justified assumptions, without any clinical trial having been carried out on the combination itself.

The Board validated this approach, noting that clinical modeling based on experimental data and scientifically justified assumptions “is an established approach for predicting the efficacy of anti-HCV drugs” (point 3.16), and that the application, combined with the in vitro data available for each compound taken individually, contained elements going beyond mere verbal statements, establishing a mechanism and a technical concept supporting the claimed therapeutic use (points 3.13 and 3.18).

The opponent also disputed whether this third-party document could be taken into account, arguing that it did not form part of the common general knowledge of the skilled person. The Board rejected this objection: since the document was cross-referenced in the application itself, it was irrelevant whether or not it belonged to the common general knowledge; the cross-reference was sufficient, in this case, to incorporate it into the disclosure for the purposes of sufficiency of disclosure.

The Board thus applied, more than twenty years later, the principle laid down in T 609/02 in almost the same terms: sufficiency must be satisfied on the basis of the information provided in the application as filed and the common general knowledge available at the time; additional evidence filed later may only confirm, never establish by itself, sufficiency (point 3.1). The post-published document relating to the marketing authorization of the medicinal product Maviret was therefore used only as confirmatory evidence (point 3.19).

T 265/23 thus shows that the credibility of the therapeutic effect may be sufficiently supported by in silico modeling, without it being necessary to produce clinical data on the combination itself, provided that the model is based on an already established methodology and on real experimental data for each component of the invention and for their combination.

What Article 56 EPC requires

To be inventive before the EPO, a claim must demonstrate an improved or unexpected technical advantage over the prior art, generally supported by data. The EPO does not apply a principle of structural non-obviousness: a new molecule is not inventive merely because it has a new structure. The structural difference over the closest prior art must produce an advantageous property, such as affinity, stability or therapeutic activity.

These data may be filed with the application or produced later, during examination or opposition proceedings, as post-filed data. No restriction applies to data filed at the filing date. For post-filed data, however, the technical effect they demonstrate must be encompassed by the technical teaching of the original application and embodied by the invention as originally disclosed: this is the principle laid down by the Enlarged Board of Appeal in G 2/21.

T 265/23: inventive step tested against in silico evidence

The same case, T 265/23, illustrates this framework, this time on the ground of inventive step under Article 56 EPC. At the priority date, the two claimed compounds were still at an early stage of development: they were not mentioned in a review published around the same time on emerging HCV therapies. The prior art provided only in vitro data on each compound taken individually, without any teaching on treatment duration or on the benefits of a possible treatment without interferon or ribavirin. The question before the Board was therefore whether the skilled person, faced with this prior art, would have had a reasonable expectation of success in obtaining a viable combined treatment without interferon or ribavirin. Yet, in the patent at issue, these advantages were established only by the predictive clinical simulation model.

The opponent attempted to turn this argument to its advantage at the inventive step stage, arguing that the same predictive modeling, described as an “established” approach, should have led the skilled person, as a matter of routine, to the claimed combination on the basis of the available in vitro data alone. The Board rejected this argument (points 5.22 and 5.23): the specific model used to make the predictions did not form part of the skilled person’s common general knowledge at the priority date and was therefore not accessible without prior knowledge of the application itself. The opponent had also not shown that another accessible model or algorithm would have made it possible to arrive at the same prediction.

This point deserves emphasis: the fact that a modeling methodology had already been used and published elsewhere, which contributes to the credibility of the technical effect under Article 83 EPC, is not sufficient to make it common general knowledge accessible to the skilled person for the purposes of Article 56 EPC. These two requirements do not fully overlap: the model must be sufficiently established and validated to convince the Board as to sufficiency of disclosure, while remaining sufficiently specific and not widely disseminated so as not to render the claimed solution obvious. It is these two qualities, credibility on the one hand and specificity on the other, that made it possible, in this same case, to acknowledge both sufficiency of disclosure and inventive step.

In summary

The EPO does not yet have a uniform approach to the patentability of in silico inventions. The reception appears favorable for structural features, for certain well-characterized functional interactions and, as T 265/23 shows, even for the therapeutic effect itself, provided that the modeling is based on a seriously founded and validated methodology.

What T 265/23 shows is not that in silico data are merely added to experimental data, but that they may replace the data that would be the most costly to obtain, namely clinical trial data, provided that the model itself is based on real data, such as in vitro data for each component, and on scientifically justified assumptions.

An insufficiency present at filing remains irremediable, whatever the nature of the missing evidence. The most prudent approach therefore remains to base the invention on in silico modeling that is as robust as possible, supported where appropriate by bench-generated experimental data.

Calibrating this foundation, that is, deciding what, in a computational model, can reasonably stand in for bench data, and how far a functional or therapeutic claim can rely on it without crossing the line into insufficiency, requires bringing together, from the drafting stage of the application, biotech patent expertise and a detailed understanding of the computational model or algorithm underlying the in silico invention.

This is Plasseraud IP’s dual positioning: a healthcare team composed of patent engineers specialized in the pharmaceutical and biotech fields, together with a team dedicated to artificial intelligence capable of assessing the robustness of a model and the credibility of the results it produces. This combination makes it possible to support both the construction of a case relying on in silico data and its defense, or challenge, before the EPO.