COVID-19: Protection of IP for AI tools
ArtemisDiana / Shutterstock.com
AI and computer-related inventions that use biological data are hot property, but what IP is best to protect them? Lauris Kemp, Claire Irvine, and Susan Keston of HGF explain.
Bioinformatics is a buzzword at the moment. The ability of modern computing systems to process complex biological data has meant healthcare and biotech is changing.
Bioinformatics covers a vast swathe of technology which applies maths and computing to biological data.
The number of bioinformatics applications being filed is increasing year on year as can be seen from figure 1 below which shows just one IPC class for bioinformatics.
Figure 1: PCT Applications published under G16B IPC code
If we look at what is in these applications, they include using maths to remove sequencing noise from single cell data; maths being used to allow the better comparison between microbiome samples with variable reads in each sample; mathematical models of the brain to in silico test Alzheimer’s drugs; and database processing methods to filter metagenomic databases faster.
Looking to more complex AI inventions, these have given rise to, for example, patent applications dealing with AI to detect early cancer biomarkers. Patent applications have also been filed to training an AI system to detect which antigens are expressed on the surface of tumour cells to allow targeted T cells to be made.
Also, we cannot leave the realm of AI without mentioning patent applications to AlphaFold: a game changer in terms of drug discovery which uses AI to determine protein structure.
Why protect with a patent?
A question often asked is: why protect this kind of invention when it could be kept secret?
The answer to this is that patent protection can nevertheless be very useful. If we use the example of a sequencing company. Providing the best sequencing results is why your customers come to you.
If you cannot protect this method of sequencing then your competitors can use the same method (as keeping it secret does not prevent them from independently developing and exploiting your method) and you have lost your hold over the market. Being able to demonstrate you can protect your market share is also important for investment.
The same can be said for other types of businesses which are not service providers (like the sequencing company example above). If we go back to one of the first examples above in the introduction: of T cell vaccines.
Your ability to provide the best T cell vaccines comes from your bioinformatics pipeline and being able to select which antigens are on the surface of the cancer. If you do not protect this, then your competitor can arrive at the same algorithm to select the antigens on the surface of the tumour cell, and can provide the same treatment – the activated T cells, as you.
Therefore, protecting bioinformatics pipelines is important for many types of business where the competitive edge comes from clever bioinformatics. Otherwise, the competition can develop and exploit the same and you could be left without any legal instrument (i.e. a patent) to prevent them from doing so.
Not only does keeping your bioinformatics secret not protect your place in the market, but it also does not protect against your competitors filing their own patent application if they do arrive at your pipeline independently. This could then be enforced against you upon grant.
It may be useful to think about bioinformatics methods in a similar way to industrial downstream processing methods for food tech. These are in-house methods which could potentially be kept secret. Yet these are protected via patents which are used against competitors all the time in European Patent Office (EPO) opposition and national litigation proceedings.
Bioinformatics methods are no different from these in-house manufacturing methods. The important thing to think about instead is if the bioinformatics is important for your business and if someone else independently arrived at your method, what would that mean to your business?
So you have an invention in this space, you have decided you want to protect it via patent protection to stay ahead of the competition but how do you go about getting protection for it?
The good news is the EPO is incredibly friendly to these sorts of inventions. The EPO examiners who examine these types of applications often have bioinformatics degrees and PhDs. They understand the science very well.
As to the rules on patenting this kind of invention, the EPC does not allow patenting of mathematical methods “as such” (Article 52 EPC). This “as such” is the important part and means the exclusion is interpreted narrowly and is called “the first hurdle” in the “two hurdle approach” applied to bioinformatics inventions.
To get over this hurdle, you need to make the claim not simply an abstract mathematical method. The simplest way of doing this is to specify the method is a “computer-implemented” one. The reference to hardware in the claim overcomes this hurdle.
The second hurdle is generally the trickier one. This hurdle is assessed under inventive step. If we take the example of a claim to a method to provide more accurate DNA sequencing, the EPO examiner will look at this claim and find everything that helps achieve the practical end result i.e. the more accurate DNA sequencing. If the maths steps as recited in the claim are not specific enough in the detail of how they get to this more accurate DNA sequencing, these will not be seen as helping achieve that practical end result and the examiner will not include the math steps in the assessment of inventive step.
This can clearly be a problem with bioinformatics inventions as generally the inventive step lies in the maths steps. If these are not taken into account for inventive step then what remains will likely be found obvious. It is therefore important to have back-ups in the description with more details of the maths/data processing.
Another point on inventive step arises specifically for AI inventions. Simply applying generic AI to a problem will not get you inventive step. There has to be something further. This could be the way that the training data is pre-processed which is not the norm in the field.
This could be the specific way in which the AI is set up to process the data (a particular algorithm not typically applied to the problem in hand). The devil is very much in the detail and a robust assessment of the prospects of success for a given invention by a multi-disciplinary attorney team can mean the difference between grant and refusal of an application. The EPO have released specific guidelines on AI which can be found here.
Turning back to general considerations for bioinformatics applications, the objections most often raised by the bioinformatics examiners at the EPO relate to clarity and inventive step (specifically the second hurdle).
With regards to clarity, it is often the case that non-standard terms are used given the evolving field, and no definitions are supplied in the application. The EPO is very strict on clarity and a definitions section for the terms in the claim is always useful.
With regards to inventive step (the second hurdle) it is often the case the applicant gets stuck between a rock and a hard place. The EPO examiner may have raised an inventive step objection that more specific maths steps are required otherwise they will not include the clever maths in the inventive step analyses as explained above.
But to include the specific maths, there is no reasonable back-up positions from the general maths in claim 1 other than a specific example. If we were to amend the claim based on the specific example, then the whole example is likely required to be added into claim 1 (as picking and choosing parts of an example to add to a claim is generally against EPO added matter rules).
The resulting claim will be far too specific to be commercially useful to your client. Therefore, nested back-up positions describing the maths in claim 1 in increasingly more detail in the description are very useful as are multi-dependent claims. This approach means that the applicant is more likely to achieve commercially useful protection at grant.
Next, let’s think about what types of claims can be included in a claim set. If the method is mathematical analyses rather than AI for example, we may have a claim set with a method to the analyses and further independent claims to a computer program and a computer-readable medium to implement the method.
In terms of claims for AI inventions, possible claim types are shown below in Figure 2.
Figure 2: A claim set for an AI invention
With reference to the numbers in Figure 2 above, independent claims can be to:
- A method of training an AI system, for example with a specific type of new and non-obvious training data
- A trained AI system
- A method of using the trained system to process real world data, eg if the system was trained to detect biomarker patterns for early cancer diagnosis, the real world data will be a sample from a patient who needs assessed
- The output of the trained system, e.g. if the trained system models small molecules fitting into an enzyme binding pocket, the output will be a new small molecule
Lastly, it is not only patent protection for general methods that provides value for bioinformatics companies. There is immense value in your databases too. These can be protected by trade secrets and a trade secret policy in place in your business can demonstrate this value in a tangible way to investors. In this way IP protection for bioinformatics companies is different from normal “wet science” biotech.
In summary, when thinking about inventions in the bioinformatics/ AI space there is often a balance of considerations to be made to determine the best protection for any particular part of a bioinformatics pipeline as summed up by the see-saw in Figure 3 below. Patent protection is just one path for consideration but the balance may well tip in its favour.
Figure 3: Getting the balance right for each bioinformatics asset in your business: Patents or trade secrets?
Lauris Kemp is a patent director at HGF. She can be contacted at: firstname.lastname@example.org
Claire Irvine is a partner and patent attorney at HGF. She can be contacted at: email@example.com
Susan Keston is a partner and patent attorney at HGF. She can be contacted at: firstname.lastname@example.org
Bioinformatics, AI, data, healthcare, computing, patents, trade secrets, European Patent Office, inventive step
COVID-19: Protection of IP for AI tools