In recent years, artificial intelligence (AI) has gained a ubiquitous presence in most technological fields, and it comes as no surprise that AI tools are also gaining popularity in efforts to accelerate drug discovery.

With pharmaceutical companies already experimenting with this new technology at preclinical stages to generate candidate active compounds, many are actively trying to protect their results with patents directed to those compounds and the AI tools used to generate them.

On the other hand, IP protection can only be as strong as the science behind it. Whether a drug candidate is developed using AI-assistance or through human endeavour (or a combination of the two), if an invention is a mere modification of what is already known or is not backed up with any (or sufficient) data showing the resultant compounds have the claimed activity, it may be very difficult or impossible to obtain valid patent protection.

Care should be taken not to rely solely on AI-generated data to support the critical subject-matter claimed in a patent application. Whilst AI and machine learning models can process large sets of data to arrive at calculated conclusions, they do not guarantee that that the resultant chemical structures generated will be new or have the desired pharmaceutical activity – and this can result in objections to claims to those compounds by the patent office, as discussed below.

A quick look at the patent filings by AI-assisted drug discovery companies reveals that applicants attempting to patent AI-assisted drugs are facing some fairly consistent challenges. Applications relating to chemical and pharmaceutical inventions are often being deemed to lack plausibility due to claims encompassing a vast number of in-silico chemical compounds within any claimed broader structure. This was the case for Recursion Pharmaceutical’s PCT application (WO2024039689) relating to Heterocycle RBM39 Modulators. Claim 1 of this application encompassed many thousands of chemical compounds and resulted in a non-establishment of opinion on patentability by the relevant patent office, as it was deemed implausible that even a fraction of those molecules would have the claimed activity (in the absence of any data proving otherwise). Other applications claiming large numbers of potential active compounds have also suffered the same “plausibility” objection, as well as lack of novelty and inventive step. Applicants should therefore carefully consider the scope of their claims in light of these decisions, balancing the need to protect the area around lead compounds that are shown to work whilst avoiding overly ambitious claiming that will lead to rejection of the claims.

It is therefore important to file any patent application for the correct molecule or group of molecules for which there is at least some efficacy data (whether in vitro or, preferably, in vivo). As well as the “plausibility” issues, filing too early with just in silico data may also mean that when more effective molecular structures in the same area are further elucidated and supporting data is generated, the previous applications may become citable prior art, reducing the likelihood of being able to obtain protection for the actual useful structures.

Another group of key patent-filers in the AI-assisted drug discovery field are AI drug discovery platform companies, as there is often highly valuable IP associated with the platform technology itself. AI-assisted platforms generally outshine non-AI computational software, by using large quantities of high-quality data and are trained to predict much more accurately parameters like clinical efficacy, toxicity, and manufacturability.

Patent protection may be useful for AI platform companies due to the market advantages patents may provide in seeking investors, establishing company image, and easing day-to-day business operations (e.g., disclosing IP to collaborators). However, the disadvantages of the fast-moving nature of the field appear to affect platform companies more than drug companies. For example, new algorithmic approaches may become irrelevant in the three or four years it takes for a patent to grant, or even in the few months it takes to draft a patent application. Once granted, patents in this field may also prove to be difficult to enforce as it may not be certain whether a competitor is using the patented invention. Furthermore, patent applications directed to new algorithms for drug-discovery may fall foul of the effective moratorium of patents for software in many countries.

In many cases, it may therefore be desirable for AI-platform companies to consider using the trade secrets approach to protect their inventions. Even this approach is not without problems, as successful algorithms may be reverse engineered, and the fluid and competitive nature of the labour market means that it can be challenging to maintain a successful trade secret policy.

Overall, AI-assisted drug discovery companies are currently having a tough time protecting their inventions, but it is very early days in terms of the development of technologies underpinning this field and as case law and the technologies develop further, we should see more clarity in what can and can’t be patented, both in terms of AI-assisted platforms and the drug candidates produced through their use. Given the potential commercial prizes that can be gained from obtaining patent protection for early-stage technologies in any field, patent protection for AI-assisted drug discovery should not be dismissed lightly, even in light of the current environment.

If you would like to discuss whether your AI-assisted drug discovery technology might be protectable by patents or any other IP right, please get in touch and one of our expert attorneys will be able to assist.