27/02/2020
Drug discovery is expensive. Computers are an important tool in combating this, because their computations can reduce the number of time-consuming physical tests needed. The use of computers in drug discovery is the subject of a great deal of research and we saw an example of this in the news last week when it was reported that a powerful new antibiotic had been discovered using artificial intelligence (see J. Stokes et al., “A Deep Learning Approach to Antibiotic Discovery”, Cell, vol. 180, no. 4, pp. 688-702.e13, 2020. Available: 10.1016/j.cell.2020.01.021, widely reported by the media).
Given the investment made in computer processes to assist in drug discovery, we can expect owners of any inventions to want to protect their investment using the patent system. But, patent applications in this field face particular difficulties due to the restrictions on patenting inventions that relate to excluded subject-matter, such as computer programs and mathematical methods. This is a problem when a new drug discovery process relies on advances made in computer programs and mathematical methods.
The news of this antibiotic presents a timely case study to look at how the pharmaceutical industry might protect its investment in any new computer-assisted drug discovery process.
Excluded subject-matter at the EPO
The European Patent Office (EPO) will not grant a patent to an invention that relates solely to any of a number of categories of excluded subject-matter, including computer programs and mathematical methods.
Even if the invention as a whole is more than ‘just’ a computer program, any steps or features that relate solely to one of these categories of excluded subject-matter will be ignored when assessing inventive step, unless they form part of a ‘technical’ solution to a ‘technical’ problem.
Where the invention lies in the use of a computer program or a mathematical method we need to look closely at how it interacts with the claimed system or method as a whole. This includes what effect this feature has on its output.
Could the AI-assisted drug discovery process be patentable at the EPO?
The article reports that the process of obtaining the antibiotic included the following steps:
- The researchers trained a neural network to predict molecules with antibacterial activity through optimisation over a training set, including 2,000 compounds and results of each compound’s effectiveness in inhibiting bacterial growth.
- Once trained, the researchers applied the neural network to a library of around 6,000 compounds under various stages of investigation for human diseases, identifying compounds predicted to have antibacterial activity.
- The researchers then identified candidate compounds for further study by prioritizing compounds among those predicted to have antibacterial activity that are structurally dissimilar to existing antibiotics.
Regardless of the EPO’s excluded subject-matter restrictions, it is unlikely that a patent could be obtained for the basic idea of using a neural network to assist the antibiotic discovery process because it is known to use a neural network to assist in drug discovery. So the mere use of the neural network will not confer inventive step.
The step of prioritizing compounds that are structurally dissimilar to existing antibiotics looks more promising. This step appears to solve a problem that arises in the context of antibiotics and so might not be obvious from any general teaching of neural networks in drug discovery. But the prioritizing step is a step of ordering a list of candidates on a computer and might not be ‘technical’ by itself. We must be able to demonstrate that this step forms part of a technical solution to a technical problem. If not, the EPO could simply ignore it when assessing inventive step. One way to demonstrate this is if the step has some physical effect in the real world, as opposed to producing results that are confined to data in a computer. This is because the real world is generally viewed as ‘technical’ under EPO practice.
The prioritizing step provides an advantage in that it can identify compounds having antibacterial activity and to which bacteria might not have evolved resistance. If the patent claim includes a step of obtaining a physical sample of the identified antibiotic, then the prioritizing step has a physical effect in the real world. This prioritizing step affects which compound is selected to prepare the physical sample. In other words, the step of obtaining the physical sample has the effect of ‘anchoring’ the computer-based prioritizing step in the real world.
But many applicants would prefer a patent claim that does not include a step of obtaining the physical sample of the identified antibiotic. They might prefer a patent claim that ended with a step of outputting a candidate compound on a computer screen. Getting a patent at the EPO for such a claim is more difficult, but could still be possible. The issue is whether a process that outputs the identity of a candidate compound, but does not provide a physical sample of that compound is itself ‘technical’.
The same issue has been faced by engineers who increasingly use computer modelling to simulate physics as part of a design or development process. This is typical of work performed by modern engineers and the trend of recent case law in Europe has been to accept that this work is technical in itself, essentially by virtue of it being the work carried out engineers, who are concerned with ‘technical’ matters almost by definition.
But the situation is still fluid despite this trend in case law because question has been referred to the EPO’s Enlarged Board of Appeal, i.e. the highest appeal body at the EPO. The Enlarged Board is currently deciding whether computer simulations are patentable by themselves or whether a final ‘anchoring’ step is needed. This referral is discussed in more detail in our previous article on the subject.
The Enlarged Board is yet to issue a final decision so the situation is not confirmed either way. In the meantime it is important to file patent applications that are flexible enough to account for the uncertainty in this field.
Conclusion
Computer technologies such as AI are being used in fields far away from their computer science origins. In a few years’ time, drug discovery teams might look very different, most likely combining people with expertise in the life sciences and in electronics and computer science. We at Reddie and Grose, have formed a specialist interdisciplinary group with patent attorneys from backgrounds across these disciplines to assist our clients working at the cutting edge of technology where these fields converge. If you agree with us that AI and other computer technologies are going to become a big part of drug discovery in the coming years and would like to discuss how we can help you secure patents in this area, don’t hesitate to contact us.
This article is for general information only. Its content is not a statement of the law on any subject and does not constitute advice. Please contact Reddie & Grose LLP for advice before taking any action in reliance on it.
