AI has been hailed as a potentially revolutionary tool for accelerating and enhancing the difficult and expensive process of drug discovery. Medicine perhaps represents the field in which AI has the most to offer humanity. However, the nascent field of AI-assisted drug design also highlights the need for IP strategy to be as forward-looking and innovative as the science it seeks to protect. Firstly, strong IP protection for the clinical drug candidate will be of paramount importance, regardless of how the drug was discovered. For AI drug discovery platform companies there may also be highly valuable IP associated with the platform technology itself. An important question for such companies is when, and even if, to file patents for this IP. Finally, the life-blood of any company seeking to innovate in the pharmaceutical field is high quality data, and AI companies will be no exception. A successful IP strategy will have to balance each of these priorities. 

Protecting core product IP is paramount

Taking a drug product through clinical trials to launch takes many years and hundreds of millions of dollars of investment. As such, strong core IP covering the marketed drug product is generally regarded as essential, and the same will apply to AI-derived drugs. Patent protection for a new molecule entering clinical trials is how innovator pharmaceutical companies seek to ensure a return of investment for the millions of dollars required to take the drug through to market. For AI companies focused on drug discovery and design, a primary focus of the patent strategy should therefore be on ensuring strong core patent protection for the clinical candidate. As such, the patentability of a drug candidate must be a key component in drug design. Furthermore, the strength of the IP protecting a drug product will only be as strong as the science behind it. If your drug candidate is a mere obvious modification of a known product you will face an uphill battle convincing the patent office that you have a patentable invention, regardless of whether it was derived using AI-tool or through human endeavour. 

Alfie (a.k.a. “AlphaKat”)

AI-drug discovery companies will also have to take care not to file too early for a drug candidate. The filing of a patent application starts the clock ticking on patent expiry and loss of exclusivity for a drug product. Additionally, the case law on sufficiency of disclosure with respect to in silico data and therapeutic effect is still developing. It is possible that the expected tsunami of chemical structure predictions from AI may even lead to an increase in the enablement requirement for a new chemical entity (IPKat). Furthermore, the original molecule identified by an AI algorithm is likely to require further optimization before it reaches the clinic, for example to take account of toxicity, manufacturability and in vivo efficacy. The final clinical candidate may therefore be different to that originally identified by the AI algorithm in silico. If the patent application was filed too early it may not therefore cover the eventual clinical lead and worse, may become citable prior art against a subsequent filing. As patent law currently stands, pursuing a patent based solely on AI-modelling data would be a brave decision indeed. 

A quick look at the patent filings of AI-assisted drug companies reveals that some may still be on a learning curve as far as developing and protecting core IP is concerned. Recursion Therapeutics is a Salt Lake city based company focused on AI assisted phenotypic screening. On its website, Recursion promises to “leverage technology to reshape the typical drug discovery funnel by broadening the number of potential therapeutic starting points beyond hypothesized and human-biased targets“. However, whilst Recursion has multiple AI-optimized small molecule compounds in the clinic, a number of these compounds (REC-2282 and REC-4881) were known and patented by other parties, requiring Recursion to take a licence. Recursion also appears to be struggling in the prosecution of its own patent applications for its drug compounds. In one instance, the ISA has found the compounds defined by the pending claims as lacking in novelty, one of several straightforward possibilities, and/or not supported by any experimental data (WO 2024/039689). These examples are a good illustration of the principle that AI-assistance does not guarantee AI-invention (or patentability) (IPKat). 

Protecting the platform: To patent or trade secret? 

For companies seeking to build a platform for AI drug discovery, IP protection for the technology itself will also be of key importance. A critical question for such companies is whether the core technology should (and can) be protected by patents or whether trade secrets are more appropriate. Companies in the drug discovery field have to balance the benefits of patent protection for their platform technologies with the risks of disclosing the details of their technology to competitors. 

A primary reason for filing patents protecting the drug discovery platform may merely be to attract and reassure investors that the company has a concrete technological offering, and that this is protected by a strong metaphorical business “moat”. Additionally, having IP on file makes discussions with potential collaborators and partners safer and easier to conduct. 

However, there are also important potential downsides to patenting AI platform technology. The first of these is the fast moving nature of the field. A new algorithmic approach 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 and file the patent application. Another downside to a solely patent-orientated approach to protecting platform IP is the time and resources associated with drafting and prosecuting a patent application. The complicated nature of AI technology means that any patent draft is likely to need a heavy degree of input from the inventors. Time taken assisting with patent drafts may thus consume valuable researcher time at the expense of further innovation. Finally, even if a patent is efficiently drafted and prosecuted, and is still relevant to a company’s technology when granted, the resulting patent may still be challenging to enforce. It can be difficult to know if a competitor is using your patented algorithmic approach behind their API. 

Given all the potential pitfalls of a patent-orientated approach, AI-drug discovery companies may therefore be tempted to protect their technology as trade secrets. A trade secret approach has the advantage that the details of the technology never have to be disclosed. However, a trade secret approach also has its disadvantages. The competitive nature of the AI-assisted drug discovery field may mean that if a company’s approach is particularly successful, it is highly likely to be reverse engineered by competitors. In such cases, trade secrets provide no protection. The success of OpenAI’s LLMs (i.e. ChatGPT) is a canonical example of how quickly fast-follows can reproduce a technology once the general approach is known. We await to see how and if OpenAI will choose to enforce its patents in the field (IPKat). Furthermore, maintaining a successful trade secret policy can itself represent a significant challenge, especially in fluid and competitive labour markets such as AI. 

The decision whether to patent the AI platform technology must therefore be made on a case-by-case basis for each technological advance and type of company. Given the competitive and fast-paced nature of the field, a thoughtful and focused patent strategy would appear to offer the greatest advantage, at least for truly innovative companies with their sights on building a field defining AI-assisted drug discovery platform. 

Currently, there is a diversity of approach within the field. Some of the more established companies have little-or-no published patent applications relating to their platform, whilst other companies have taken a more aggressive patent strategy. Relay Therapeutics, a AI drug-design company focused on the optimisation of chemical structures by modelling molecular dynamics, has multiple products in the clinic but no platform patent applications filed in its name. Google DeepMind, by contrast, has many pending patent applications covering the different aspects and modules of AlphaFold, the core technology behind its AI-assisted drug discovery spin-out Isomorphic Labs

Final thoughts

It is tempting to think of AI-assisted drug discovery companies as just another category of platform company for drug development. However, there is a key distinction. Unlike other drug discovery platforms, effective AI-assisted drug discovery will be heavily reliant on vast quantities of high quality data. A machine learning model for accurately predicting the clinical efficacy, toxicity and manufacturability of a new drug will need a solid foundation of training data. At the moment, such data are primarily in the hands of large pharmaceutical companies. The IP strategy, and the associated commercialisation model of AI-assisted drug discovery companies, will therefore need to take account of the need for access to this data. This potentially places a huge bargaining chip in the hands of big pharma looking to partner with AI-drug discovery companies. In negotiations with the pharma industry, the ability for an AI platform company to demonstrate a strong IP position supported by a meaningful and effective patent strategy protecting truly innovative technology will likely bear the most fruit. 

Further reading





Source link

Leave a Comment

Your email address will not be published. Required fields are marked *